11 micro-topics in AI Systems, each with the evidence that proves you have it and the written reason behind every link. Before you start, it rests on 8 other domains. Downstream, it holds up 1 domain.
11 topics · depth 10–13 of 16 · 11 internal links · 10 in · 2 out
Load-bearing links first. Each one says what in this domain rests on what outside it, and why.
3 links
Run an agent loop rests on Execute and return tool results
The loop calls tools.
Decide when a step needs an agent rests on Draw the model boundary
The agent question is the model boundary drawn one level up.
Run an agent loop rests on Recover from tool failure
Loops depend on surviving individual tool failures.
1 link
Manage working memory rests on Carry structured state instead of transcripts
Working memory is structured state.
1 link
Run an agent loop rests on Route requests in FastAPI
The tools it calls have to be served.
1 link
Bound loops with budgets rests on Enforce spend caps in code
Spend caps are the same mechanism at the loop level.
1 link
Reflect on results rests on Write deterministic checks
Self-evaluation borrows from how you score output.
1 link
Checkpoint and resume agent state rests on Checkpoint long-running jobs
Resumability is the same pattern as in batch jobs.
1 link
Insert human approval rests on Maintain audit trails
Approvals are only meaningful if they are recorded.
1 link
Persist long-term memory rests on Store and query vectors in Postgres
Semantic recall is usually vector retrieval.
In prerequisite order. Each idea names the artifact that closes it, the market demand that put it on the map, and what it stands on.
Depth 10 · conceptual · scaffolded · M20 §20.1
Cycles perceive, decide, act and observe
Execute and return tool results · Tool Use & MCP
The loop calls tools.
Route requests in FastAPI · FastAPI
The tools it calls have to be served.
Recover from tool failure · Tool Use & MCP
Loops depend on surviving individual tool failures.
Depth 11 · conceptual · scaffolded · M20 §20.1
Decomposes a goal into ordered steps
Planning organizes iterations of a loop that runs.
Depth 11 · conceptual · guided · M20 §20.1
Evaluates its own output and retries
Reflection evaluates the last iteration.
Write deterministic checks · LLM Evals
Self-evaluation borrows from how you score output.
Depth 11 · conceptual · scaffolded · M20 §20.4
Caps iterations, tokens and wall time
A ceiling bounds a loop.
Enforce spend caps in code · Model Integration
Spend caps are the same mechanism at the loop level.
Depth 11 · conceptual · scaffolded · M20 §20.3
Holds task state across steps
Working state persists across iterations.
Carry structured state instead of transcripts · Context Engineering
Working memory is structured state.
Depth 11 · conceptual · scaffolded · M20 §20.5
Pauses for a person at consequential steps
A pause interrupts a loop.
Approval and budget are the two ways a loop is bounded.
Maintain audit trails · AI Safety & Governance
Approvals are only meaningful if they are recorded.
Depth 12 · conceptual · guided · M20 §20.3
Stores and retrieves across sessions
Persistence extends working memory.
Store and query vectors in Postgres · Embeddings & Vectors
Semantic recall is usually vector retrieval.
Depth 12 · representational · guided · M20 §20.4
Makes transitions explicit rather than emergent
A graph makes a plan's transitions explicit.
Depth 12 · meta · guided · M20 §20.4
Uses an agent only where a step chooses its own next step
Draw the model boundary · Tool Use & MCP
The agent question is the model boundary drawn one level up.
You need to have built one to know when not to.
Depth 13 · conceptual · guided · M20 §20.4
Persists position so a run recovers
Model an agent as a state graph
You checkpoint at defined states.
Checkpoint long-running jobs · Data Ingestion
Resumability is the same pattern as in batch jobs.
Depth 13 · conceptual · guided · M20 §20.5
Specializes roles and defines handoffs
Model an agent as a state graph
Handoffs are transitions between agents.
Everything downstream that names an idea here as a prerequisite, grouped by where it lives.
2 links
Distinguish a harness from an execution layer rests on Run an agent loop
A harness wraps a loop that runs.
Set retry and budget ceilings rests on Bound loops with budgets
The ceiling moves the loop's budget into the harness.
Retool. Retrain. Relaunch.
375 ideas. 17 weeks. No tuition, ever.
Vets Who Code is a veteran-run 501(c)(3). The accelerator is free, remote, and we don’t take a share of your first paycheck.
Free · Remote · 17 weeks · EIN 86-2122804