← Journey home

14 Sep 2026 → 13 Sep 2027

The adaptive schedule.

AI changes too quickly for a fixed 365-day syllabus. Week 1 is our active curriculum: we will build an AI agent using a local Ollama model. At the end of every week, we review what we learned, what changed in the field, and what evidence we produced before committing to the next week. Later weeks show direction, not a locked curriculum.

Mon–Wed · 90 min Read, build, verify, and recordThursday · 90 min Repair, practise, and explainFriday · 2 hours Integrate and ship evidenceSaturday · 45 min Curate and shareSunday · 30 min Reflect and plan

Our working product: Subra AI — a local Ollama agent

Our first working product is a small JavaScript agent that runs against a local Ollama model through its OpenAI-compatible API. It receives a goal, follows explicit instructions, can choose one allowlisted local tool, and reports the result for human review.

First acceptance check: We can run the agent locally, stream its response, observe its model and token behaviour, and verify one bounded tool call without sending prompts to a cloud model.

Week 1 is intentionally small: one local model, one agent loop, and one safe tool. System prompts guide behaviour but are not treated as a security boundary; application code validates tool names, arguments, limits, and approval requirements.

A chapter explains planned work. A session result records what we actually ran and learned. A milestone is complete only when its linked acceptance evidence supports that status.

Week 1

D1–7Local AI agent foundation
Weekly outcome

A working AI agent powered entirely by a local Ollama model.

Readiness check

We need an editor and one available assistant; Day 1 includes a browser fallback for JavaScript.

Agent team

Our local Ollama agent performs the bounded task; a review assistant may critique the design, but we inspect every prompt, tool request, and result.

Human approval gate

We can explain the difference between a prompted chatbot and an agent, then demonstrate that only an allowlisted tool with validated arguments can run.

Decision we record

We choose one model that fits our hardware, a practical context size, and one tightly bounded tool for the first agent loop.

Proof we produce

Ollama runtime notes, normal and streamed SDK calls, sampling experiments, agent source code, tests, and a recorded local demo

Thursday practice · 20 min

We spend 20 minutes on one small JavaScript/TypeScript problem: first attempt, then agent critique and one test.

Weekend
Sat · We spend 45 minutes curating Ollama runtime notes, normal and streamed SDK calls, sampling experiments, agent source code, tests, and a recorded local demo. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 2

D8–14Local AI agent foundation · Provisional
Weekly outcome

One usable coding workflow with a measured local-model option.

Readiness check

We build on Week 1: A working AI agent powered entirely by a local Ollama model. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A setup assistant helps configure the chosen tool; a benchmark reviewer checks results from the same tasks.

Human approval gate

We explain tool permissions and reproduce one task; if local inference is unsuitable, we document the available hosted fallback.

Decision we record

We choose one coding assistant and one model using our hardware, cost, and review needs.

Proof we produce

tool setup record and a three-task comparison with actual timings

Thursday practice · 20 min

We spend 20 minutes explaining this week’s design tradeoff aloud, including a failure case and a simple diagram.

Weekend
Sat · We spend 45 minutes curating tool setup record and a three-task comparison with actual timings. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 3

D15–21Local AI agent foundation · Provisional
Weekly outcome

A minimal React Native entry screen with a reviewed development loop.

Readiness check

We build on Week 2: One usable coding workflow with a measured local-model option. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A coding agent scaffolds the screen; a separate review task checks validation; a test assistant proposes edge cases.

Human approval gate

We can create and list a local entry, explain the state changes, and show that empty input is handled.

Decision we record

We choose the smallest app shell and one local entry shape for the companion.

Proof we produce

mobile screen recording, entry validation checks, and a reviewed PR

Thursday practice · 20 min

We spend 20 minutes reviewing one small diff or preparing a mentoring explanation; peer participation is optional and recorded honestly.

Weekend
Sat · We spend 45 minutes curating mobile screen recording, entry validation checks, and a reviewed PR. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 4

D22–28Local AI agent foundation · Provisional
Weekly outcome

A first evidence milestone with a draft-only coordinator workflow.

Readiness check

We build on Week 3: A minimal React Native entry screen with a reviewed development loop. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

Hermes or our existing assistant prepares evidence drafts; a privacy reviewer checks only the selected artifacts.

Human approval gate

We trace each draft claim to a real artifact and explicitly approve any public update.

Decision we record

We choose a coordinator and keep manual publishing as the initial fallback.

Proof we produce

reviewed weekly draft, permission record, and mobile demo link

Thursday practice · 20 min

We spend 20 minutes comparing one role requirement with our evidence and revising a short public-profile or case-study draft.

Weekend
Sat · We spend 45 minutes curating reviewed weekly draft, permission record, and mobile demo link. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 5

D29–35TypeScript backend systems · Provisional
Weekly outcome

The mobile app reads a sample entry from a Node.js API.

Readiness check

We build on Week 4: A first evidence milestone with a draft-only coordinator workflow. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A backend agent drafts one endpoint; a mobile agent connects it; a reviewer checks response handling.

Human approval gate

We demonstrate both a successful request and a useful mobile error when the API is stopped.

Decision we record

We choose one Node framework and keep the API small enough to explain.

Proof we produce

one API contract, mobile-to-API recording, and request log

Thursday practice · 20 min

We spend 20 minutes on one small JavaScript/TypeScript problem: first attempt, then agent critique and one test.

Weekend
Sat · We spend 45 minutes curating one API contract, mobile-to-API recording, and request log. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 6

D36–42TypeScript backend systems · Provisional
Weekly outcome

A mobile create-and-list flow with runtime input validation.

Readiness check

We build on Week 5: The mobile app reads a sample entry from a Node.js API. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

An API agent implements the contract; a test agent checks bad inputs; a mobile reviewer checks duplicate taps.

Human approval gate

We create an entry from mobile and reject malformed API input without a server crash.

Decision we record

We define entry fields, request validation, and predictable HTTP errors.

Proof we produce

create/list demo, API examples, and invalid-input checks

Thursday practice · 20 min

We spend 20 minutes explaining this week’s design tradeoff aloud, including a failure case and a simple diagram.

Weekend
Sat · We spend 45 minutes curating create/list demo, API examples, and invalid-input checks. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 7

D43–49TypeScript backend systems · Provisional
Weekly outcome

Journal entries survive an API restart in PostgreSQL.

Readiness check

We build on Week 6: A mobile create-and-list flow with runtime input validation. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A data agent proposes the schema; a migration reviewer checks changes; a test agent checks saved entries.

Human approval gate

We back up sample data, apply the migration, restart the API, and retrieve the same entry.

Decision we record

We choose one query approach and a reversible first schema migration.

Proof we produce

entry schema, migration, persistence test, and restart recording

Thursday practice · 20 min

We spend 20 minutes reviewing one small diff or preparing a mentoring explanation; peer participation is optional and recorded honestly.

Weekend
Sat · We spend 45 minutes curating entry schema, migration, persistence test, and restart recording. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 8

D50–56TypeScript backend systems · Provisional
Weekly outcome

Two test users cannot read each other’s journal entries.

Readiness check

We build on Week 7: Journal entries survive an API restart in PostgreSQL. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

An identity assistant explains the integration; a coding agent adds ownership; a security reviewer tests denial.

Human approval gate

We verify user B cannot read or change user A’s entry, including direct API requests.

Decision we record

We use a maintained identity integration and define entry ownership.

Proof we produce

auth diagram, ownership tests, and two-user mobile check

Thursday practice · 20 min

We spend 20 minutes comparing one role requirement with our evidence and revising a short public-profile or case-study draft.

Weekend
Sat · We spend 45 minutes curating auth diagram, ownership tests, and two-user mobile check. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 9

D57–63TypeScript backend systems · Provisional
Weekly outcome

A consolidated mobile/API foundation with one reviewed design decision.

Readiness check

We build on Week 8: Two test users cannot read each other’s journal entries. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A debugging agent reproduces the highest-priority defect; a design critic questions the API choice.

Human approval gate

We rerun create, persistence, and ownership checks before accepting any extra API surface.

Decision we record

We decide whether the current screen needs GraphQL; REST remains valid.

Proof we produce

closed foundation gaps, updated tests, and a REST/GraphQL decision note

Thursday practice · 20 min

We spend 20 minutes on one small JavaScript/TypeScript problem: first attempt, then agent critique and one test.

Weekend
Sat · We spend 45 minutes curating closed foundation gaps, updated tests, and a REST/GraphQL decision note. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 10

D64–70TypeScript backend systems · Provisional
Weekly outcome

One repeatable background job prepares an entry export.

Readiness check

We build on Week 9: A consolidated mobile/API foundation with one reviewed design decision. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A worker agent implements one handler; a failure-test agent resubmits the same job; a reviewer checks limits.

Human approval gate

We submit the same job twice and verify one logical export plus a visible failure state.

Decision we record

We choose a queue, job key, and bounded retry policy; caching is deferred.

Proof we produce

export job, duplicate-job test, and mobile status recording

Thursday practice · 20 min

We spend 20 minutes explaining this week’s design tradeoff aloud, including a failure case and a simple diagram.

Weekend
Sat · We spend 45 minutes curating export job, duplicate-job test, and mobile status recording. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 11

D71–77TypeScript backend systems · Provisional
Weekly outcome

The existing companion runs reproducibly in a local service stack.

Readiness check

We build on Week 10: One repeatable background job prepares an entry export. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A container assistant drafts configuration; a release reviewer checks the startup guide and existing tests.

Human approval gate

We follow the guide from a clean checkout and demonstrate mobile create, persistence, and access denial.

Decision we record

We define a small milestone checklist and defer features that do not pass it.

Proof we produce

container setup, smoke checks, and a foundation milestone report

Thursday practice · 20 min

We spend 20 minutes reviewing one small diff or preparing a mentoring explanation; peer participation is optional and recorded honestly.

Weekend
Sat · We spend 45 minutes curating container setup, smoke checks, and a foundation milestone report. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 12

D78–84Python AI services · Provisional
Weekly outcome

A tested Python text utility we can explain from TypeScript knowledge.

Readiness check

We build on Week 11: The existing companion runs reproducibly in a local service stack. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A Python assistant explains syntax; a reviewer asks us to trace the function; a test agent suggests edge cases.

Human approval gate

We run the utility and tests and explain its input, output, and error behaviour.

Decision we record

We use one environment and package workflow with a small typed function.

Proof we produce

Python utility, three tests, and a language comparison note

Thursday practice · 20 min

We spend 20 minutes comparing one role requirement with our evidence and revising a short public-profile or case-study draft.

Weekend
Sat · We spend 45 minutes curating Python utility, three tests, and a language comparison note. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 13

D85–91Python AI services · Provisional
Weekly outcome

One bounded FastAPI text endpoint with a tested contract.

Readiness check

We build on Week 12: A tested Python text utility we can explain from TypeScript knowledge. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A Python API agent wraps the utility; a contract reviewer compares request and response examples.

Human approval gate

We prove valid input returns the documented result and malformed input receives a predictable error.

Decision we record

We define the text-in/result-out boundary; Node retains product persistence.

Proof we produce

FastAPI endpoint, request schema, and invalid-input tests

Thursday practice · 20 min

We spend 20 minutes on one small JavaScript/TypeScript problem: first attempt, then agent critique and one test.

Weekend
Sat · We spend 45 minutes curating FastAPI endpoint, request schema, and invalid-input tests. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 14

D92–98Python AI services · Provisional
Weekly outcome

An AI-service adapter with bounded external calls and no duplicate product database.

Readiness check

We build on Week 13: One bounded FastAPI text endpoint with a tested contract. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

An HTTP-client agent drafts the adapter; a failure reviewer checks timeout behaviour and data minimization.

Human approval gate

We prove a slow upstream call stops within the configured limit and that no full journal database is copied into Python.

Decision we record

We keep the Python service stateless until a specific persistence requirement appears.

Proof we produce

adapter tests, timeout demonstration, and data-ownership decision

Thursday practice · 20 min

We spend 20 minutes explaining this week’s design tradeoff aloud, including a failure case and a simple diagram.

Weekend
Sat · We spend 45 minutes curating adapter tests, timeout demonstration, and data-ownership decision. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 15

D99–105Python AI services · Provisional
Weekly outcome

One model generates a draft summary from supplied journal text.

Readiness check

We build on Week 14: An AI-service adapter with bounded external calls and no duplicate product database. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A model adapter agent connects the provider; an output reviewer checks the summary against the supplied text.

Human approval gate

We reject invalid output and manually check three summaries for unsupported claims.

Decision we record

We select one available provider and validate a small summary schema.

Proof we produce

summary endpoint, malformed-output test, and recorded sample outputs

Thursday practice · 20 min

We spend 20 minutes reviewing one small diff or preparing a mentoring explanation; peer participation is optional and recorded honestly.

Weekend
Sat · We spend 45 minutes curating summary endpoint, malformed-output test, and recorded sample outputs. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 16

D106–112Python AI services · Provisional
Weekly outcome

A small repeatable summary evaluation and initial model traces.

Readiness check

We build on Week 15: One model generates a draft summary from supplied journal text. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A dataset assistant drafts cases; an evaluation reviewer challenges expected answers; a trace assistant records metadata.

Human approval gate

We label the cases ourselves, inspect failures, and confirm traces exclude private raw text by default.

Decision we record

We start with human labels and deterministic checks before automated judging.

Proof we produce

ten-case dataset, baseline results, and one redacted model trace

Thursday practice · 20 min

We spend 20 minutes comparing one role requirement with our evidence and revising a short public-profile or case-study draft.

Weekend
Sat · We spend 45 minutes curating ten-case dataset, baseline results, and one redacted model trace. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 17

D113–119Python AI services · Provisional
Weekly outcome

Mobile users request, inspect, and accept a draft summary.

Readiness check

We build on Week 16: A small repeatable summary evaluation and initial model traces. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

An integration agent connects Node to Python; a mobile agent adds the draft UI; a reviewer checks acceptance semantics.

Human approval gate

We show that a draft is not saved as an accepted update without our action, and Python failure leaves the entry usable.

Decision we record

We keep summaries as drafts until accepted, with Node owning saved results.

Proof we produce

mobile summary flow, service failure test, and second milestone report

Thursday practice · 20 min

We spend 20 minutes on one small JavaScript/TypeScript problem: first attempt, then agent critique and one test.

Weekend
Sat · We spend 45 minutes curating mobile summary flow, service failure test, and second milestone report. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 18

D120–126Azure production foundations · Provisional
Weekly outcome

A costed, minimal Azure deployment plan for the existing services.

Readiness check

We build on Week 17: Mobile users request, inspect, and accept a draft summary. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A cloud planning assistant prepares options; a cost reviewer checks assumptions and non-free resources.

Human approval gate

We approve chargeable resources and understand that a budget alert is not a spending cap.

Decision we record

We choose one region and compute option; Container Apps, App Service, and AKS are compared on paper.

Proof we produce

resource plan, estimate, budget alert, and teardown checklist

Thursday practice · 20 min

We spend 20 minutes explaining this week’s design tradeoff aloud, including a failure case and a simple diagram.

Weekend
Sat · We spend 45 minutes curating resource plan, estimate, budget alert, and teardown checklist. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 19

D127–133Azure production foundations · Provisional
Weekly outcome

A tested cloud identity and secret-access boundary.

Readiness check

We build on Week 18: A costed, minimal Azure deployment plan for the existing services. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

An identity assistant proposes access scopes; a security reviewer checks excess permissions.

Human approval gate

We verify the intended identity can access one secret or resource and an unauthorized identity cannot.

Decision we record

We choose a workload identity and narrowly scoped resource access.

Proof we produce

identity diagram, one successful access check, and one denied check

Thursday practice · 20 min

We spend 20 minutes reviewing one small diff or preparing a mentoring explanation; peer participation is optional and recorded honestly.

Weekend
Sat · We spend 45 minutes curating identity diagram, one successful access check, and one denied check. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 20

D134–140Azure production foundations · Provisional
Weekly outcome

A small Azure deployment reachable from the mobile companion.

Readiness check

We build on Week 19: A tested cloud identity and secret-access boundary. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A deployment agent prepares configuration; an access reviewer checks ingress; a release assistant runs smoke checks.

Human approval gate

We verify mobile-to-Node access, protected Node-to-Python access, and a rollback to the previous revision.

Decision we record

We set low scaling limits and keep Python reachable only through the intended service path.

Proof we produce

deployment configuration, mobile smoke test, and rollback record

Thursday practice · 20 min

We spend 20 minutes comparing one role requirement with our evidence and revising a short public-profile or case-study draft.

Weekend
Sat · We spend 45 minutes curating deployment configuration, mobile smoke test, and rollback record. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 21

D141–147Azure production foundations · Provisional
Weekly outcome

A single observable request across the service boundary.

Readiness check

We build on Week 20: A small Azure deployment reachable from the mobile companion. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

An instrumentation agent adds correlation; a privacy reviewer checks logs; an incident assistant traces one failure.

Human approval gate

We follow one mobile request through both services and prove the alert fires for the chosen failure.

Decision we record

We choose trace fields, redaction rules, and one actionable error alert.

Proof we produce

correlated request trace, forced-error alert, and diagnosis note

Thursday practice · 20 min

We spend 20 minutes on one small JavaScript/TypeScript problem: first attempt, then agent critique and one test.

Weekend
Sat · We spend 45 minutes curating correlated request trace, forced-error alert, and diagnosis note. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 22

D148–154Azure production foundations · Provisional
Weekly outcome

One reproducible infrastructure change and a reviewed deployment pipeline.

Readiness check

We build on Week 21: A single observable request across the service boundary. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

An infrastructure agent drafts one resource; a plan reviewer checks replacement risk; a CI agent adds one deployment stage.

Human approval gate

We inspect the plan for unexpected deletion or replacement and confirm deployment can be withheld without affecting CI tests.

Decision we record

We choose Bicep or Terraform and require approval before deployment changes.

Proof we produce

one IaC resource, reviewed plan, and pipeline run

Thursday practice · 20 min

We spend 20 minutes explaining this week’s design tradeoff aloud, including a failure case and a simple diagram.

Weekend
Sat · We spend 45 minutes curating one IaC resource, reviewed plan, and pipeline run. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 23

D155–161Azure production foundations · Provisional
Weekly outcome

A consolidated cloud milestone with one recovery exercise.

Readiness check

We build on Week 22: One reproducible infrastructure change and a reviewed deployment pipeline. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A security reviewer audits the current path; a recovery assistant drafts a runbook; a release reviewer checks gaps.

Human approval gate

We demonstrate recovery from one chosen failure and accurately label any untested or undeployed path.

Decision we record

We record remaining cloud risks and the next most valuable fix.

Proof we produce

cloud checklist, one recovery drill, and unresolved-risk log

Thursday practice · 20 min

We spend 20 minutes reviewing one small diff or preparing a mentoring explanation; peer participation is optional and recorded honestly.

Weekend
Sat · We spend 45 minutes curating cloud checklist, one recovery drill, and unresolved-risk log. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 24

D162–168Reliable LLM applications · Provisional
Weekly outcome

A justified model choice for our existing summary task.

Readiness check

We build on Week 23: A consolidated cloud milestone with one recovery exercise. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A benchmark agent runs the existing cases; a reviewer manually checks output differences.

Human approval gate

We compare identical inputs and disclose unavailable providers or missing cost data.

Decision we record

We choose one model on quality, privacy, latency, and measured cost.

Proof we produce

same-dataset model comparison and selection record

Thursday practice · 20 min

We spend 20 minutes comparing one role requirement with our evidence and revising a short public-profile or case-study draft.

Weekend
Sat · We spend 45 minutes curating same-dataset model comparison and selection record. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 25

D169–175Reliable LLM applications · Provisional
Weekly outcome

One versioned prompt improvement with a small regression gate.

Readiness check

We build on Week 24: A justified model choice for our existing summary task. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A prompt agent proposes a change; an adversarial reviewer supplies conflicting input; an evaluation agent reruns cases.

Human approval gate

We inspect the same evaluation cases and reject unsupported improvements or new critical failures.

Decision we record

We define which output failures block a prompt change.

Proof we produce

prompt versions, attack examples, and before/after evaluation

Thursday practice · 20 min

We spend 20 minutes on one small JavaScript/TypeScript problem: first attempt, then agent critique and one test.

Weekend
Sat · We spend 45 minutes curating prompt versions, attack examples, and before/after evaluation. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 26

D176–182Reliable LLM applications · Provisional
Weekly outcome

Search finds relevant entries in a small owned corpus.

Readiness check

We build on Week 25: One versioned prompt improvement with a small regression gate. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

An embedding assistant explains similarity; a search agent drafts the query; a relevance reviewer checks labels.

Human approval gate

We retrieve only the signed-in user’s entries and inspect the five query results manually.

Decision we record

We choose a local retrieval baseline; Azure AI Search remains a later comparison.

Proof we produce

five labelled queries and a first search result list

Thursday practice · 20 min

We spend 20 minutes explaining this week’s design tradeoff aloud, including a failure case and a simple diagram.

Weekend
Sat · We spend 45 minutes curating five labelled queries and a first search result list. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 27

D183–189Reliable LLM applications · Provisional
Weekly outcome

Changed journal entries update their searchable chunks reliably.

Readiness check

We build on Week 26: Search finds relevant entries in a small owned corpus. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

An ingestion agent splits text; a lineage reviewer checks source IDs; a test agent changes and deletes entries.

Human approval gate

We trace each result to an entry and prove a deleted entry is no longer returned.

Decision we record

We choose one text format, chunk rule, and source identifier.

Proof we produce

chunk examples, update/delete tests, and source mapping

Thursday practice · 20 min

We spend 20 minutes reviewing one small diff or preparing a mentoring explanation; peer participation is optional and recorded honestly.

Weekend
Sat · We spend 45 minutes curating chunk examples, update/delete tests, and source mapping. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 28

D190–196Reliable LLM applications · Provisional
Weekly outcome

One mobile answer cites retrieved evidence or says it cannot answer.

Readiness check

We build on Week 27: Changed journal entries update their searchable chunks reliably. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A retrieval agent assembles context; a generation agent drafts the answer; a reviewer checks citations against source text.

Human approval gate

We can open every cited entry and verify that an unsupported question produces an explicit no-answer result.

Decision we record

We define a minimal retrieval-and-answer contract and citation format.

Proof we produce

cited mobile answer, no-answer case, and source-opening demo

Thursday practice · 20 min

We spend 20 minutes comparing one role requirement with our evidence and revising a short public-profile or case-study draft.

Weekend
Sat · We spend 45 minutes curating cited mobile answer, no-answer case, and source-opening demo. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 29

D197–203Reliable LLM applications · Provisional
Weekly outcome

A consolidated RAG baseline with at most one justified retrieval improvement.

Readiness check

We build on Week 28: One mobile answer cites retrieved evidence or says it cannot answer. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A failure analyst groups bad results; an experiment agent changes one variable; a reviewer checks the comparison.

Human approval gate

We use the same queries before and after the experiment and keep the baseline if the change is not useful.

Decision we record

We choose hybrid search, reranking, or no change from observed failures.

Proof we produce

failure review and one controlled retrieval experiment

Thursday practice · 20 min

We spend 20 minutes on one small JavaScript/TypeScript problem: first attempt, then agent critique and one test.

Weekend
Sat · We spend 45 minutes curating failure review and one controlled retrieval experiment. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 30

D204–210Reliable LLM applications · Provisional
Weekly outcome

A small RAG evaluation report calibrated against our own labels.

Readiness check

We build on Week 29: A consolidated RAG baseline with at most one justified retrieval improvement. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A dataset agent proposes cases; an evaluator agent runs one metric; a human-label reviewer checks disagreements.

Human approval gate

We inspect disagreements and never substitute an automated score for source verification.

Decision we record

We choose one RAGAS or custom metric and keep human review in the release gate.

Proof we produce

ten question/source pairs, evaluator comparison, and regression result

Thursday practice · 20 min

We spend 20 minutes explaining this week’s design tradeoff aloud, including a failure case and a simple diagram.

Weekend
Sat · We spend 45 minutes curating ten question/source pairs, evaluator comparison, and regression result. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 31

D211–217Reliable LLM applications · Provisional
Weekly outcome

A reviewed RAG milestone with measurable limits.

Readiness check

We build on Week 30: A small RAG evaluation report calibrated against our own labels. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A release assistant gathers existing results; a security reviewer checks the retrieval boundary; a load assistant runs one bounded test.

Human approval gate

We pass cross-user denial and citation checks and label the actual deployment and evaluation status.

Decision we record

We set release criteria using actual evaluation and observed failure behaviour.

Proof we produce

RAG milestone report, mobile demo, access tests, and request timing

Thursday practice · 20 min

We spend 20 minutes reviewing one small diff or preparing a mentoring explanation; peer participation is optional and recorded honestly.

Weekend
Sat · We spend 45 minutes curating RAG milestone report, mobile demo, access tests, and request timing. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 32

D218–224Agents and MCP · Provisional
Weekly outcome

A deterministic workflow prepares a proposed weekly update.

Readiness check

We build on Week 31: A reviewed RAG milestone with measurable limits. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A workflow agent assembles selected entries; a reviewer checks each drafted claim against those entries.

Human approval gate

We show that preparing a draft performs no public write and every claim has a source.

Decision we record

We define where fixed steps are sufficient and where model judgment is needed.

Proof we produce

workflow diagram, draft output, and approval boundary

Thursday practice · 20 min

We spend 20 minutes comparing one role requirement with our evidence and revising a short public-profile or case-study draft.

Weekend
Sat · We spend 45 minutes curating workflow diagram, draft output, and approval boundary. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 33

D225–231Agents and MCP · Provisional
Weekly outcome

One bounded agent selects a read-only evidence tool.

Readiness check

We build on Week 32: A deterministic workflow prepares a proposed weekly update. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A tool agent drafts the schema; a policy reviewer checks allowed inputs; an adversarial agent tries an out-of-scope call.

Human approval gate

We reject calls outside the user’s scope and stop execution at the configured call limit.

Decision we record

We allow one narrow tool and a small call budget before adding writes.

Proof we produce

tool schema, permitted and denied calls, and execution trace

Thursday practice · 20 min

We spend 20 minutes on one small JavaScript/TypeScript problem: first attempt, then agent critique and one test.

Weekend
Sat · We spend 45 minutes curating tool schema, permitted and denied calls, and execution trace. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 34

D232–238Agents and MCP · Provisional
Weekly outcome

An existing local agent can read one approved capability through MCP.

Readiness check

We build on Week 33: One bounded agent selects a read-only evidence tool. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A protocol assistant explains MCP; a server agent wraps the existing read tool; a reviewer checks workspace scope.

Human approval gate

We expose only selected sample evidence and verify the server cannot read unrelated paths.

Decision we record

We start with a local transport and postpone remote authentication until needed.

Proof we produce

local MCP server, tool-call trace, and denied-access case

Thursday practice · 20 min

We spend 20 minutes explaining this week’s design tradeoff aloud, including a failure case and a simple diagram.

Weekend
Sat · We spend 45 minutes curating local MCP server, tool-call trace, and denied-access case. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 35

D239–245Agents and MCP · Provisional
Weekly outcome

An approval pause can resume without duplicating the accepted action.

Readiness check

We build on Week 34: An existing local agent can read one approved capability through MCP. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A state-machine agent drafts transitions; a recovery reviewer interrupts and resumes the run.

Human approval gate

We prove resume requires approval and repeating it does not repeat the side effect.

Decision we record

We choose an explicit state machine or LangGraph for one workflow.

Proof we produce

state diagram, pause/resume test, and duplicate-resume check

Thursday practice · 20 min

We spend 20 minutes reviewing one small diff or preparing a mentoring explanation; peer participation is optional and recorded honestly.

Weekend
Sat · We spend 45 minutes curating state diagram, pause/resume test, and duplicate-resume check. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 36

D246–252Agents and MCP · Provisional
Weekly outcome

Users can inspect and delete one remembered preference.

Readiness check

We build on Week 35: An approval pause can resume without duplicating the accepted action. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A memory assistant suggests the minimum field; a privacy reviewer checks scope; a mobile agent adds user controls.

Human approval gate

We demonstrate consent, inspection, and deletion and prove another user cannot retrieve the preference.

Decision we record

We distinguish a saved preference from unrestricted long-term agent memory.

Proof we produce

preference UI, access tests, and deletion result

Thursday practice · 20 min

We spend 20 minutes comparing one role requirement with our evidence and revising a short public-profile or case-study draft.

Weekend
Sat · We spend 45 minutes curating preference UI, access tests, and deletion result. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 37

D253–259Agents and MCP · Provisional
Weekly outcome

A measured delivery exercise using scoped specialist agents.

Readiness check

We build on Week 36: Users can inspect and delete one remembered preference. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A planner, coder, independent reviewer, and test assistant collaborate on one existing defect; a docs agent drafts the log.

Human approval gate

We reproduce the defect, validate the fix ourselves, and resolve disagreements before merging.

Decision we record

We keep roles that improve review quality and remove unnecessary orchestration.

Proof we produce

agent task briefs, reviewed defect, time/cost notes, and completed fix

Thursday practice · 20 min

We spend 20 minutes on one small JavaScript/TypeScript problem: first attempt, then agent critique and one test.

Weekend
Sat · We spend 45 minutes curating agent task briefs, reviewed defect, time/cost notes, and completed fix. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 38

D260–266Agents and MCP · Provisional
Weekly outcome

A consolidated agent flow with a small safety regression set.

Readiness check

We build on Week 37: A measured delivery exercise using scoped specialist agents. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

An adversarial assistant proposes three attacks; a trace reviewer checks tool paths; a release agent gathers results.

Human approval gate

We verify that missing approval, an out-of-scope request, and a call-budget breach cannot execute the intended action.

Decision we record

We choose blocking checks for scope, approval, and stopping behaviour.

Proof we produce

agent regression results, selected traces, and unresolved-failure log

Thursday practice · 20 min

We spend 20 minutes explaining this week’s design tradeoff aloud, including a failure case and a simple diagram.

Weekend
Sat · We spend 45 minutes curating agent regression results, selected traces, and unresolved-failure log. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 39

D267–273Mobile product hardening · Provisional
Weekly outcome

The existing mobile AI view streams and cancels predictably.

Readiness check

We build on Week 38: A consolidated agent flow with a small safety regression set. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A streaming agent connects the transport; a mobile reviewer checks state transitions; a test assistant interrupts the connection.

Human approval gate

We cancel a response, confirm work stops where supported, and prevent stale chunks from updating a new request.

Decision we record

We define partial-response and cancellation behaviour before selecting transport.

Proof we produce

stream/cancel recording and interrupted-connection checks

Thursday practice · 20 min

We spend 20 minutes reviewing one small diff or preparing a mentoring explanation; peer participation is optional and recorded honestly.

Weekend
Sat · We spend 45 minutes curating stream/cancel recording and interrupted-connection checks. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 40

D274–280Mobile product hardening · Provisional
Weekly outcome

A scoped on-device feasibility result for the companion.

Readiness check

We build on Week 39: The existing mobile AI view streams and cancels predictably. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A mobile inference assistant configures one candidate; a benchmark reviewer checks device conditions and limitations.

Human approval gate

We record the real device, latency, memory, and quality; an unsuitable model produces a documented no-go result.

Decision we record

We choose whether a small local capability earns its device and maintenance cost.

Proof we produce

one-device benchmark, comparison, and go/no-go decision

Thursday practice · 20 min

We spend 20 minutes comparing one role requirement with our evidence and revising a short public-profile or case-study draft.

Weekend
Sat · We spend 45 minutes curating one-device benchmark, comparison, and go/no-go decision. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 41

D281–287Mobile product hardening · Provisional
Weekly outcome

One controlled attachment flow, or a completed core backlog item.

Readiness check

We build on Week 40: A scoped on-device feasibility result for the companion. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

An upload agent implements one type; a security reviewer checks validation; a mobile agent shows progress and errors.

Human approval gate

We reject an invalid or oversized file and verify the owner can delete an accepted attachment.

Decision we record

We choose one attachment type and explicit size and retention limits.

Proof we produce

one upload demo, invalid-file checks, or a documented deferred-feature decision

Thursday practice · 20 min

We spend 20 minutes on one small JavaScript/TypeScript problem: first attempt, then agent critique and one test.

Weekend
Sat · We spend 45 minutes curating one upload demo, invalid-file checks, or a documented deferred-feature decision. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 42

D288–294Mobile product hardening · Provisional
Weekly outcome

A reviewed architecture proposal for one actual coupling problem.

Readiness check

We build on Week 41: One controlled attachment flow, or a completed core backlog item. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

An architecture assistant maps dependencies; a critic challenges the options; a documentation assistant records feedback.

Human approval gate

We defend the recommendation ourselves and distinguish actual peer feedback from an AI simulation.

Decision we record

We keep a modular design unless a measured need justifies extraction.

Proof we produce

architecture diagram, short RFC, and review response

Thursday practice · 20 min

We spend 20 minutes explaining this week’s design tradeoff aloud, including a failure case and a simple diagram.

Weekend
Sat · We spend 45 minutes curating architecture diagram, short RFC, and review response. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 43

D295–301Mobile product hardening · Provisional
Weekly outcome

One critical workflow recovers from duplicate or interrupted work.

Readiness check

We build on Week 42: A reviewed architecture proposal for one actual coupling problem. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A failure agent reproduces the case; an implementation agent fixes it; a reviewer checks duplicate execution.

Human approval gate

We repeat the same failure and show the logical result remains consistent.

Decision we record

We choose an idempotency or outbox fix based on a reproduced failure.

Proof we produce

failure reproduction, recovery test, and consistency explanation

Thursday practice · 20 min

We spend 20 minutes reviewing one small diff or preparing a mentoring explanation; peer participation is optional and recorded honestly.

Weekend
Sat · We spend 45 minutes curating failure reproduction, recovery test, and consistency explanation. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 44

D302–308Mobile product hardening · Provisional
Weekly outcome

One measured performance improvement on the mobile-to-service path.

Readiness check

We build on Week 43: One critical workflow recovers from duplicate or interrupted work. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A profiling assistant records measurements; a performance reviewer checks the experiment; a coding agent makes one change.

Human approval gate

We rerun identical inputs and device conditions and keep the change only if the result supports it.

Decision we record

We optimize the observed bottleneck and compare under the same conditions.

Proof we produce

baseline profile, change, and repeated measurement

Thursday practice · 20 min

We spend 20 minutes comparing one role requirement with our evidence and revising a short public-profile or case-study draft.

Weekend
Sat · We spend 45 minutes curating baseline profile, change, and repeated measurement. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 45

D309–315Mobile product hardening · Provisional
Weekly outcome

A rehearsed response to one user-visible incident.

Readiness check

We build on Week 44: One measured performance improvement on the mobile-to-service path. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

An incident assistant organizes observations; a telemetry reviewer checks the alert; a documentation agent drafts the postmortem.

Human approval gate

We diagnose and recover the chosen failure using the runbook and record actual elapsed times.

Decision we record

We select one service objective, alert, owner, and recovery action.

Proof we produce

runbook, incident timeline, and follow-up fix

Thursday practice · 20 min

We spend 20 minutes on one small JavaScript/TypeScript problem: first attempt, then agent critique and one test.

Weekend
Sat · We spend 45 minutes curating runbook, incident timeline, and follow-up fix. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 46

D316–322Mobile product hardening · Provisional
Weekly outcome

A reviewed security boundary across the actual mobile and AI flow.

Readiness check

We build on Week 45: A rehearsed response to one user-visible incident. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

Mobile, API, and AI review tasks inspect their respective boundaries; a validation agent reproduces one finding.

Human approval gate

We verify sensitive mobile data handling, entry ownership, and approval enforcement before the milestone.

Decision we record

We prioritize validated findings and explicitly record remaining risks.

Proof we produce

focused threat update, reproduced finding, fix, and verification

Thursday practice · 20 min

We spend 20 minutes explaining this week’s design tradeoff aloud, including a failure case and a simple diagram.

Weekend
Sat · We spend 45 minutes curating focused threat update, reproduced finding, fix, and verification. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 47

D323–329Mobile product hardening · Provisional
Weekly outcome

A consolidated product candidate and a documented handover exercise.

Readiness check

We build on Week 46: A reviewed security boundary across the actual mobile and AI flow. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A release assistant identifies gaps; a documentation agent prepares the handover; a reviewer tries the instructions.

Human approval gate

We explain the system without generated answers and label simulated feedback honestly; no Staff-level impact is inferred from a solo exercise.

Decision we record

We freeze milestone scope and separate individual work from demonstrated collaboration.

Proof we produce

handover guide, peer or simulated feedback, and milestone gap list

Thursday practice · 20 min

We spend 20 minutes reviewing one small diff or preparing a mentoring explanation; peer participation is optional and recorded honestly.

Weekend
Sat · We spend 45 minutes curating handover guide, peer or simulated feedback, and milestone gap list. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 48

D330–336Career evidence and applications · Provisional
Weekly outcome

A fourth milestone reviewed against existing acceptance evidence.

Readiness check

We build on Week 47: A consolidated product candidate and a documented handover exercise. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A release reviewer gathers tests and evaluations; an evidence agent links artifacts; a human makes the release decision.

Human approval gate

We distinguish local verification, test deployment, and production operation and do not declare unfinished checks passed.

Decision we record

We release, defer, or reduce scope based on actual results.

Proof we produce

milestone report, mobile demonstration, deployed/local status, and limitations

Thursday practice · 20 min

We spend 20 minutes comparing one role requirement with our evidence and revising a short public-profile or case-study draft.

Weekend
Sat · We spend 45 minutes curating milestone report, mobile demonstration, deployed/local status, and limitations. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 49

D337–343Career evidence and applications · Provisional
Weekly outcome

A clear portfolio narrative backed by the year’s work.

Readiness check

We build on Week 48: A fourth milestone reviewed against existing acceptance evidence. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A portfolio editor drafts from artifacts; a claim reviewer checks every metric; a recruiter critic tests clarity.

Human approval gate

We verify claims and distinguish professional impact, personal projects, and simulated exercises.

Decision we record

We position for roles matching our evidence and label Staff roles as scope-dependent stretch targets.

Proof we produce

updated resume, one case study, and portfolio feedback

Thursday practice · 20 min

We spend 20 minutes on one small JavaScript/TypeScript problem: first attempt, then agent critique and one test.

Weekend
Sat · We spend 45 minutes curating updated resume, one case study, and portfolio feedback. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 50

D344–350Career evidence and applications · Provisional
Weekly outcome

A coding-interview checkpoint informed by recurring practice.

Readiness check

We build on Week 49: A clear portfolio narrative backed by the year’s work. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A mock interviewer asks questions; a hint assistant intervenes only on request; a reviewer critiques after the attempt.

Human approval gate

We explain the solution, edge cases, and complexity ourselves before reading the critique.

Decision we record

We target the weakest pattern in our practice log rather than survey every algorithm.

Proof we produce

one mock, corrected solution, and practice comparison

Thursday practice · 20 min

We spend 20 minutes explaining this week’s design tradeoff aloud, including a failure case and a simple diagram.

Weekend
Sat · We spend 45 minutes curating one mock, corrected solution, and practice comparison. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 51

D351–357Career evidence and applications · Provisional
Weekly outcome

A system-design interview grounded in our actual product decisions.

Readiness check

We build on Week 50: A coding-interview checkpoint informed by recurring practice. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A mock interviewer changes a requirement; a design reviewer challenges capacity and failure assumptions.

Human approval gate

We defend tradeoffs ourselves and distinguish implemented experience from a proposed design.

Decision we record

We focus on one scenario and clarify requirements before proposing infrastructure.

Proof we produce

design recording, diagram, critique, and revised explanation

Thursday practice · 20 min

We spend 20 minutes reviewing one small diff or preparing a mentoring explanation; peer participation is optional and recorded honestly.

Weekend
Sat · We spend 45 minutes curating design recording, diagram, critique, and revised explanation. We publish only reviewed results and label unfinished work.Sun · We review this week’s evidence, changes in AI tooling and practice, and our remaining gaps. We then revise and publish the next week’s curriculum; future weeks remain provisional.

Week 52

D358–364Career evidence and applications · Provisional
Weekly outcome

A measured application cycle and a plan for the next quarter.

Readiness check

We build on Week 51: A system-design interview grounded in our actual product decisions. If its required checks remain incomplete, we use this week to finish or reduce scope first.

Agent team

A role-matching assistant organizes evidence; an outreach agent drafts; a reviewer checks claims and recipients.

Human approval gate

We personally approve applications and record outcomes only when received.

Decision we record

We select roles by fit and adjust the next step from actual feedback.

Proof we produce

application tracker, reviewed messages, interview notes, and retrospective

Thursday practice · 20 min

We spend 20 minutes comparing one role requirement with our evidence and revising a short public-profile or case-study draft.

Weekend
Sat · We spend 45 minutes curating application tracker, reviewed messages, interview notes, and retrospective. We publish only reviewed results and label unfinished work.Sun · We prepare the final retrospective and use the year’s evidence to plan the next quarter.
365

Final retrospective

We review the complete year: capabilities earned, products shipped, mistakes corrected, market response, and our next professional target. The compensation target remains aspirational and depends on role fit, evidence, interview performance, location, and the market.