Blog
Practical writing about agents, automation, and software architecture.
Detected Focus
Several recent posts point to the same engineering problem: agents do not only need better prompts. They need a runtime contract around the prompt.
The minimal model
A useful harness defines the agent’s phase, allowed actions, required evidence, exit condition, and blocker rules. That is the difference between an AI that tries to help and an agent that can be operated.
An agent harness is the control layer around an AI agent. It defines allowed actions, state transitions, required evidence, and stop conditions.
Without a harness, the model decides too much from chat context. It may edit before inspecting, mark work done without running checks, keep looping after the useful work is finished, or create fake review comments because the prompt asked it to criticize.
The practical fix is to move those rules out of vibes and into an explicit contract: phase state controls which actions are allowed; evidence gates control when the agent may claim progress; exit conditions control when the loop stops; blocker rules control when the agent must ask for help; review rules prevent pointless self-critique.
Related reading
Natural-Language Agent Harnesses give a useful name to an important shift: the agent policy should be an inspectable document that a runtime executes, not invisible glue hidden inside controller code.
When an agent keeps jumping from planning to editing to testing at the wrong time, the fix is not usually another paragraph of system prompt. Put the workflow into explicit states, give each state a tiny tool policy, and make phase changes visible.
When an agent clicks, sends, pays, deletes, or extracts data, the critical truth cannot live only in model prose. Put a small evidence gate before risky tool calls: predicate, evidence type, source, decision.
The useful lesson behind Claude Code /goal is not that agents can run forever. It is that long-running agent work needs an explicit, observable exit condition: what proves done, what stays in scope, and when to stop blocked.
Open-ended instructions like “critically self-check this” accidentally reward the model for producing criticism. The fix is not less review. It is calibrated review: explicit criteria, PASS_NO_CHANGE, evidence per finding, severity thresholds, and a tiny change budget.
July 17, 2026 · AI Agent Workflows
Harness Handbook points at a practical bottleneck in agent engineering: the behavior you want to change is scattered across prompts, state managers, tool calls, policy code, and tests. Build a behavior map before editing the harness.
July 17, 2026 · AI Agent Workflows
Asking one agent to reconsider its answer often produces a more confident defense of the same mistake. A bounded challenger-and-judge loop can create real alternatives, but only if disagreement, stopping, and judge bias are engineered explicitly.
July 17, 2026 · AI Agent Workflows
An agent that writes a lesson into memory, a skill, a prompt, or its own code is deploying behavior into future runs. This guide shows how to put persistent changes through evidence, eval, approval, expiry, and rollback gates.
July 15, 2026 · AI Agent Workflows
The reliable eval system is not one automated judge. It is a closed loop that combines portable traces, deterministic invariants, narrow semantic judges, versioned production failures, adversarial tests, and human calibration.
July 13, 2026 · AI Agent Workflows
A final pass/fail score hides the part of agent work that matters most: where the run started drifting, whether it noticed, and whether it recovered. The practical replacement is a trajectory eval with checkpoints, failure labels, and recovery metrics.
July 10, 2026 · AI Agent Workflows
The useful question is moving from which model is best to what your agent harness can change, measure, persist, and roll back.
July 7, 2026 · AI Agent Security
The MOSAIC paper shifts the coding-agent security question from hostile prompts to command traces. The practical move is to audit producer-consumer state across shell commands before generated state crosses into privileged work.
July 6, 2026 · AI Agent Workflows
The serious agent pattern is no longer bigger prompts and more encouragement. It is an operating contract: measurable goal, bounded tools, context sources, verifier evidence, review notes, rollback path, and a skill update when the run teaches you something.
July 2, 2026 · AI Agent Workflows
The useful upgrade from prompt engineering is not a longer instruction block. It is a reusable loop spec: trigger, goal, allowed tools, verifier, terminal states, and memory rules. That is how repeated coding-agent work becomes operational instead of conversational.
June 26, 2026 · AI Agent Workflows
Better AI products come from improvement systems around the agent. This guide shows how to build one with deterministic checks, narrow scoring rubrics, private holdouts, calibrated judges, and promotion gates.
June 24, 2026 · AI Agent Security
Local LLM agents can touch shells, files, browsers, credentials, memory, and messaging tools. Treat their runtime layer as source code worth auditing, then turn static findings into a manual review queue instead of automatic verdicts.
June 22, 2026 · AI Agent Workflows
The useful question is not whether MCP, A2A, ACP, agents.json, Agora, ANP, LMOS, or AGNTCY wins. The useful question is which communication boundary you are designing: discovery, tool execution, task delegation, identity, transport, or runtime negotiation.
June 17, 2026 · AI Agent Workflows
Most memory evals ask whether the agent got the final answer right. MemTrace suggests a sharper unit: one durable user fact tested across age, current state, earlier state, trajectory, and contradictory evidence. That turns memory from a vague feature into a small regression suite.
June 10, 2026 · AI-first Engineering
Two 2026 papers from the same research lineage quietly retire prompt engineering as a discipline. The agent's system prompt is now a binary you can version, diff, and evolve with a 200-line loop. The four metrics that actually matter are not the ones your dashboard shows.
June 8, 2026 · AI-first Engineering
The surprising lesson from AGENTS.md benchmarks is not that context files are useless. It is that they change agent behavior, sometimes into more expensive and less useful work. Treat them as a control surface, not a repo manual.
June 3, 2026 · AI Agent Workflows
Dynamic workflows move agent work from one chat prompt into inspectable orchestration: phases, subagents, evidence, budget, permissions, adversarial review, and stop conditions. The point is not more agents. The point is better control.
May 30, 2026 · AI-first Engineering
Tests tell you whether behavior still works. Linters tell you whether code is syntactically and stylistically acceptable. An AI-slop gate catches the residue coding agents leave behind: fake comments, swallowed errors, any-casts, duplicated helpers, TODO stubs, and dead code.
May 30, 2026 · AI-first Engineering
Bad reward functions should not be treated like prompt drafts. Treat them like production incidents: preserve traces, classify the failure, patch only the implicated logic, and rerun against the same controls.
May 28, 2026 · AI-first Engineering
AI-generated interfaces often look finished before they behave correctly. A GUI playtester loop uses a separate browser agent to interact with the artifact, record screenshots and action logs, turn broken flows into reproducible bug reports, and rerun the same script after repairs.
May 28, 2026 · AI-first Engineering
Better AI coding is not mainly about better prompts. It is about the harness around the model: explicit contracts, separate builder and reviewer roles, evidence requirements, and a loop that turns failures into better specifications.
May 26, 2026 · AI Agent Workflows
The useful lesson behind Claude Code /goal is not that agents can run forever. It is that long-running agent work needs an explicit, observable exit condition: what proves done, what stays in scope, and when to stop blocked.
May 26, 2026 · AI Agent Workflows
Agent evals should not only ask whether the final answer looked good. A useful benchmark measures the whole agent system: skill routing, tool policy, evidence, outcomes, hard-fail safety cases, regressions, cost, and production drift.
May 25, 2026 · AI Agent Workflows
When an agent keeps jumping from planning to editing to testing at the wrong time, the fix is not usually another paragraph of system prompt. Put the workflow into explicit states, give each state a tiny tool policy, and make phase changes visible.
May 24, 2026 · AI-first Engineering
Natural-Language Agent Harnesses give a useful name to an important shift: the agent policy should be an inspectable document that a runtime executes, not invisible glue hidden inside controller code.
May 23, 2026 · AI-first Engineering
Long agent chats rot. A better pattern is to move decisions into small spec files, clear context between layers, and let each coding-agent session read only the artifact it needs.
May 21, 2026 · AI Agent Workflows
When an agent clicks, sends, pays, deletes, or extracts data, the critical truth cannot live only in model prose. Put a small evidence gate before risky tool calls: predicate, evidence type, source, decision.
May 21, 2026 · AI Agent Workflows
Open-ended instructions like “critically self-check this” accidentally reward the model for producing criticism. The fix is not less review. It is calibrated review: explicit criteria, PASS_NO_CHANGE, evidence per finding, severity thresholds, and a tiny change budget.
May 20, 2026 · AI-first Engineering
The arXiv survey Code as Agent Harness names the next shift in agent engineering: code is not only what agents generate. It is becoming the executable, inspectable, stateful runtime that makes agents reliable.
May 20, 2026 · AI-first Engineering
Teams do not usually start vibe coding because developers became careless. They start because onboarding is broken: docs are stale, harnesses are undocumented, system knowledge lives in people’s heads, and AI turns missing context into plausible code and Markdown.
May 19, 2026 · AI-first Engineering
A coding agent is not made reliable by one magic prompt. It needs a harness: AGENTS.md, skills, tool permissions, hooks, and evals that catch behavior drift.
May 19, 2026 · AI Agent Workflows
The useful move is not one mega assistant for all client work. Give each client project a small, isolated agent with its own memory, tasks, preview URL habit, and boring daily standup.
May 18, 2026 · AI-first Engineering
After context engineering comes decomposition: developers should stop putting everything into one prompt and instead split tasks into direct prompts, subtasks, pipelines, agent loops, or skills.
May 15, 2026 · AI-first Engineering
The next developer skill is not writing clever prompts. It is building the operating system around LLMs: data quality, model versioning, evals, guardrails, incident response, review UX, and repo instructions agents can actually follow.
May 15, 2026 · Personal AI Workflows
Voice is not good for everything. But for small agent jobs it is brutally useful: dictate a task while moving, transcribe it locally, let your existing agent handle it, and get only a short answer back.
May 12, 2026 · AI-first Engineering
In 2026, good prompting is not about one magic sentence. The better approach is to curate context, define tools and schemas, set agent rules, and verify behavior with evals.
May 11, 2026 · AI Agent Workflows
Hermes gets interesting when an agent does not only produce output, but reviews the run: execute, measure, critique, rewrite the skill, and test again. The loop pays off mainly for repeatable workflows.
April 29, 2026 · AI-first Engineering
AI-first architecture does not mean the model decides. It means AI generates options, finds risks, compresses context, and the team makes a traceable decision.
Send me a short note about your project or technical bottleneck.
Get in touch