Skip to lesson
中文
The collection/AN OPEN STUDY ROUTE

WEEK 24 / 24 · Keep what matters

Keep building

Graduation is a beginning: a new task, independent repair and an adaptable practice

10 core hours + 2 optionalREAD · BUILD · PRACTISE
Begin the lesson ↓
Keep buildingTHE PERSONAL FDE FIELD LIBRARY

The question: Can you take an unfamiliar small requirement from discovery to recovery, beyond repeating a course demonstration?

Before you begin: Aim to have a deterministic tool, a controlled agent workflow and an evidence-backed improvement experiment. State incomplete parts honestly.

FIG. 24.1THIS WEEK’S REASONING PATH
  1. 01Define a new task
  2. 02Build & verify
  3. 03Recover independently
  4. 04Choose what is next
Teaching diagram · Derived from this week’s lab; a suggested process, not observed results.

Concepts and learning objectives

Graduation is end-to-end ability rather than video or agent counts. Define a new small requirement, its inputs, permissions and success criteria. Implement, verify, use and recover it. Independently repair one small fault without code generation; documentation remains allowed.

Keep the curriculum adaptable. Requirements, data, networking, verification, permissions and maintenance are durable capabilities; models, SDKs and some control techniques change. New models may make old mechanisms unnecessary, so updates should include deletion. Treat engineering announcements as proposals to evaluate, not orders to migrate everything.

Reading and selected sections

Priority Source and selected sections Budget
Review R09 · Stanford CS146S — The Modern Software Developer, Fall 2026 — Compare current AI-assisted development topics with your gaps, without claiming degree equivalence. 25 min
Optional R46 · Anthropic — Managed Agents — Runtime and control-mechanism trade-offs; identify logic that may be removable. 35 min
Further study R40 · Berkeley — Advanced LLM Agents, Spring 2025 — Select planning, tools or verification topics from the Spring 2025 archive according to an actual bottleneck. 40 min

Studio lab

  1. Define a new requirement and non-goals yourself. Collaborate on implementation while personally reviewing data flow, permissions and important diffs. Record unverified behaviour.
  2. Perform core checks and real use, then introduce a recoverable fault. Stop, diagnose, restore and verify outstanding external effects yourself.
  3. Complete a generator-free repair or rule change and explain its cause and verification. Record remaining gaps for the next learning cycle.
  4. Prepare public work: problem, architecture, synthetic examples, verification and limitations. Distinguish actual reviewed aggregates from demonstrations. Do not invent a success story.
  5. Evaluate one model or SDK update on fixed tasks with unchanged budget and permissions. Keep the option not to upgrade. Record version, sources, adoption criteria and rollback; choose one next-month learning focus.

Deliverables and acceptance

Deliverable Evidence of completion
End-to-end new task A coherent account of requirements, implementation, verification, use and limits.
Independent recovery You take over failure without asking a model to regenerate the project.
Public work and update rules No private records; facts and demonstrations distinct; next steps follow real gaps.

Independent exercise: With the generator closed, explain a key module, repair a new small defect and recover a known failure. Mark what you cannot yet do.

Privacy and safety: Review files, build output, links, images, metadata and history before publishing. A clean secret scan is not proof of safety. Separately inspect authentication, hosting logs, analytics and permissions. External works retain their own copyright and terms.

Everyday use: Keep two or three tools that naturally fit your life and retire the other prototypes. Reduce burden rather than making every routine depend on complex automation.

Pass or revisit: Deliver, verify, recover, explain value and recognise your limits. This learning route is not a degree or employment guarantee.

A reusable collaboration prompt

Assess me through a new requirement, independent repair and recovery drill using real evidence. Prepare public synthetic demonstrations and honest limitations. Compare a model or SDK upgrade within a budget, allow no upgrade, and choose one next learning direction from demonstrated gaps.

Sources and verification scope

Reference IDs and scope follow the supplied manuscript, baseline 2026-09-27. External authenticated labs were not all completed; recheck versions and access before implementation.

  1. R09
    Stanford CS146S — The Modern Software Developer, Fall 2026 ↗

    Introduction and prerequisites in the manuscript; not a claim that all Fall materials are available.

  2. R40
    Berkeley — Advanced LLM Agents, Spring 2025 ↗

    Advanced elective; clearly labelled as a 2025 archive.

  3. R46
    Anthropic — Managed Agents ↗

    Runtime and removable control logic; not a compliance guarantee.