Authority topics
AI engineering
The engineering advantage comes from making probabilistic components legible, testable, and bounded inside deterministic systems.
Practical analysis of coding agents, agent infrastructure, context and retrieval, model/tool boundaries, verification, and the engineering systems around model output.
Core writing
Reliable AI Systems Are Shrinking the Probabilistic Surface
The strongest agent systems this week did not ask models to reason about everything. They moved explicit objectives into hard constraints, deterministic data work before context, machine interfaces into structured contracts, authority into lifecycle-bound metadata, and reliability into end-to-end workflow evidence—showing how to reserve probabilistic reasoning for ambiguity that actually needs it.
Weekly ReadThe Model Stopped Being the System
Week of June 15, 2026. One shift ran under every story: the model became a swappable component, and control moved to the layer around it — review, boundaries, context, and the people who own the result. A scan-layer TL;DR, five themes each with an operator move, one taken deep, three counter-signals, and what to track.
Weekly ReadFurther writing
Code Is Getting Cheaper. Why Are We Still Managing Delivery by Hand?
Coding agents are making execution cheaper, but software delivery still depends on people reconciling goals, plans, decisions, dependencies, code, review and release. This is the problem I am trying to move into the system with The Tam.
AI and Engineering LeadershipKISS your SOLID frontend code to keep it DRY (Part 1)
On applying SOLID, KISS, and DRY/DIE principles to frontend engineering — written in 2021 after eight years in the field, and still where most frontend complexity actually goes wrong.
Engineering practiceRecent analysis
AI Is Turning Hidden Assumptions Into System Contracts
Agent execution is getting cheap enough that requirements, acceptance criteria, autonomy boundaries, state and architecture economics can no longer remain implicit. This Weekly Read shows how to turn those hidden assumptions into inspectable system contracts using requirement recovery, executable acceptance, evidence-based handoffs, durable state and lifecycle cost models.
Weekly ReadAgent Interfaces Are Becoming Control Planes
Agent interfaces are becoming control planes: context, MCP/WebMCP actions, enterprise identity, automated feedback loops and hardware drivers now carry operational authority. This Weekly Read shows where to move provenance, delegation, retry budgets, revocation and physical interlocks before agent capability outruns the boundaries around it.
Weekly ReadThe Model Stopped Being the Interesting Variable
Week of August 17, 2026. Mistral tripled accuracy on financial filings — 26.7% to 86% — by changing how the model searches, not which model searches. Two different models from two different labs got the same roughly threefold lift from the same harness. That pattern ran through the week: authorisation splitting into four separate questions, a token optimisation that cut tokens and raised total cost by half, throughput outgrowing the review capacity around it, and execution moving into shared team channels. The leverage kept turning out to sit outside the model. A scan-layer TL;DR, five themes each with an operator move, one taken deep, three counter-signals, and what to track.
Weekly ReadCapability Outran Confirmation
Week of August 3, 2026. Frontier models generated 6,080 security patches and roughly a quarter of them worked — while a plausible-but-wrong hint in the prompt cut the success rate to one in six, and the agents followed the hint over their own evidence. The same gap opened everywhere: peak capability diverging from production reliability, agents that cannot verify themselves, operational data capping what any model can automate, and portable artifacts carrying no portable trust. Production got cheap. Confirming what was produced did not. A scan-layer TL;DR, five themes each with an operator move, one taken deep, three counter-signals, and what to track.
Weekly ReadNo Layer Was Reliable by Default
Week of July 6, 2026. The model regressed on its tools, the benchmark used to rank it turned out to be broken, and 'smarter' stopped meaning 'more reliable.' So 'one best model' quietly stopped being an architecture: a mature system now holds a portfolio, routes by verified task-fit, and checks confidence outside the model at every layer. A scan-layer TL;DR, five themes each with an operator move, one taken deep, three counter-signals, and what to track.
Weekly ReadThe System Around the Model Started to Matter More Than the Model
Week of June 8, 2026. Five themes where the value kept moving off the model and into the system around it — compute contracts, observable governance, agent runtimes, and the operating model that has to absorb the output. Plus three counter-signals worth holding in tension.
Weekly ReadOutput Got Cheap, Absorption Got Expensive
Week of June 1, 2026. The same pattern repeated across AI cost, enterprise agents, security, coding, and stablecoins: generation got cheap and the constraint moved to absorption. A scan-layer TL;DR, five themes each with an operator move, one theme taken deep, three counter-signals, and what to track.
Weekly ReadControl Planes Beat Autonomy, Tokens Stop Meaning Progress
Week of May 25, 2026. Five themes shaping how AI agents, AI coding, and AI economics are actually landing in serious engineering organizations — plus three counter-signals worth holding in tension.
Weekly ReadShort pieces
One task in my workflow can be researched in ChatGPT, implemented by Codex, challenged by Claude, and updated through MCP.
Most project tools still assume every participant is human, which leaves someone acting as the relay between chats, repositories, decisions and the board.
BuildingA generated patch is a hypothesis until someone re-runs the exploit.
Across 6,080 AI-generated patches only 26% were clean — and a plausible-but-wrong fix direction in the prompt cut success from two-thirds to one in six, with agents overriding their own contrary evidence.
AI engineeringWe keep securing the model and trusting the scaffolding around it.
Every control we put on an agent lives in the harness — so once the agent can edit its own prompts, tools and memory, a harness change is a privileged production change.
AI engineeringFour agents cleared the backlog because none was allowed to be the hero.
Astro's issue factory separated reproduction, diagnosis, verification and implementation — the improvement came from forbidding self-verification, not from adding agents.
AI engineeringAI can generate five implementations of the same rule before anyone notices the rule has split.
When generation gets cheap, duplicated business meaning gets expensive — the source of truth has to become harder to copy, not easier.
AI engineeringYour AI product is already interviewing every user — most teams throw the transcript away.
An AI interaction states intent directly instead of leaving it to be inferred from clicks, which makes conversation logs a governed product-research surface rather than discarded exhaust.
AI governanceThe strongest privacy policy may be a system that never sends the data.
On-device inference turns privacy into a property of architecture rather than paperwork — the strongest privacy policy is a system that never sends the data.
AI governanceIn agentic systems, safe defaults matter more than good warnings — the agent may never read the warning.
As agents run more workflows, security has to live in default-off execution paths, because the agent may never read the warning a human would have caught.
AI securityA good agent skill is not a clever prompt — it is organizational procedure made executable.
A reusable agent skill is organizational procedure made executable — versioned, reviewed, and owned — which is how know-how moves from tribal to institutional.
AI engineeringThe dangerous agent failure is not always a bad answer. Sometimes it is public input reaching private context.
GitLost showed a crafted public GitHub issue steering an agent with cross-repo read access into posting a private README as a public comment — no stolen credentials, just a context-separation failure. The agent runs on a service-account permission model, not a user one, so no patch closes it; the fix is architectural. If public text can steer private access, the permission model is already broken.
AI SecurityA model can get smarter and still make your system less reliable. That is what weak tool contracts do.
The newest Claude models regressed on one edit tool — inventing fields that did not match the schema, failing ~20% of the time in a real agentic session. Smarter model, worse tool behavior, only visible in a long history. The fix is not a better prompt; it is stricter schemas and runtimes that fail loudly. The tool contract is part of the product's intelligence.
AI EngineeringAI is getting good at reviewing code. That is exactly why the human's job in review has to move up.
An AI reviewer is good at the mechanical pass — but it misses the change that fits the diff and breaks the architecture. The reviewer's value moves up the abstraction stack, to the judgment and accountability the model structurally cannot supply. The mechanical reviewer can be automated; the one who owns the consequence cannot.
AI EngineeringBuilt from this thinking
Codencer
A persistent daemon that manages, executes, validates, and audits tasks performed by external coding agents.
Orchestration bridge for coding agentsRhodd
Turns architecture definitions into production-ready code, infrastructure, and CI/CD pipelines. Zero boilerplate.
AI-powered infrastructure orchestrator