Context Clues

Product strategy

Two Answers

Give a coding agent the same task twice and you get two different, plausible answers that disagree with each other. The fix is not a better model. It is a shared record of what the team already decided.

Product strategy

Greek Fire

For more than five centuries, Greek Fire's enemies could barely counter it. Then it vanished, not defeated but forgotten, because the recipe lived only in a few heads. Your systems can be lost the same way.

Product strategy

Sixteen Agents

You talk to Brief as one thing. Behind it, sixteen agents each own one part of the product picture and keep it current from your existing tools. Here is the whole team and what each one does.

Product strategy

Stigmergy

A colony coordinates with no plan and no direct messages, only the traces each worker leaves for the next. Software teams and their agents run on the same mechanism, and it works only as well as the traces do.

Product strategy

Working Memory

An agent has a large, capable working memory and, on its own, no long-term memory of your team. Bigger context windows and better retrieval improve the first and do nothing for the second.

Product strategy

Overruled Precedent

You follow the precedent set by the code around you. Almost every time it is right. This is about the one change where it is wrong, and why your agent cannot see it coming.

Product strategy

Seven Seconds

Every person leaves. Every line gets rewritten. What makes it the same company ten years on is not the people or the code. It is the memory of why, and that is the one thing nobody keeps.

Product strategy

Why coding agents ignore your rules

You wrote the rule down. The agent broke it anyway. It is not being defiant, and a longer rules file will not fix it. Here is what actually will.

Product strategy

Context, compared

Nine ways teams give coding agents context, scored against what independent research says a real solution has to do. Only one shape passes all five tests.

Product strategy

Path dependence

A wrong default caught on day one is a one-line fix. Caught on day two hundred, after everything has been built on top of it, it is a migration.

Product strategy

Goodhart's law

Your test suite was evidence that the code worked. Then the agent made passing it the goal, and evidence became the thing being manufactured.

Product strategy

Gell-Mann amnesia

The agent did not get more reliable on the second page. You lost the ability to catch it, and mistook that for the agent being right.

Product strategy

Tacit knowledge

The reason your product context never reaches the agent is not that no one wrote it down. It is that the people who hold it know more than they can tell.

Product strategy

Principal and agent

You did not hire an employee. You entered the oldest studied problem in delegation, and the AI solved the half everyone feared while leaving the half no one watched.

Product strategy

The default answer

I expected three runs of one model to diverge. They agreed almost perfectly, and the agreement turned out to be the more unsettling part.

Product strategy

The whole product

The first team proves your product is useful. The next team proves it is whole.

Product strategy

Chesterton's fence

Not seeing a reason is not the same as there being no reason, and the gap between those two is where most expensive mistakes live.

Product strategy

Survivorship bias

The most important data point is almost always the one that is missing, and it is missing for the exact reason that makes it matter.

Product strategy

Four risks

The question that dominated software for twenty years, can we build it, is the one your agent just answered. It was never the only question. It was one of four.

Product strategy

Common knowledge

The child who said the emperor was naked told nobody anything they did not already know. That is the whole point, and it is the most useful distinction I know.

AI coding agents

Ghost decisions

The agent did not make a mistake. It made a decision, one of thousands, that no one asked it to make and no one will remember, because no one made it.

Product strategy

Productivity is not value

Your engineers say AI made them faster. Your board wants to know where that shows up in the business. The honest answer is that you cannot see it, and the reason is not your dashboard.

AI coding agents

Your CLAUDE.md is a junk drawer: a migration playbook

There is no correct length for a flat context file. Trim it and the agent misses the decision it needed; fill it and the decision it needed is buried on line 2,830. The fix is not a better file.

AI coding agents

The firehose problem: your coding agent has all your permissions

Least privilege for a coding agent was never a narrower scope on the firehose. It is not handing the agent the firehose at all.

Context engineering

Loop engineering forgot the loop needs a memory

You can automate the loop. You can't automate the context the loop needs each time around, and the moment you removed yourself from every iteration, you removed the one thing that used to supply it.

AI coding agents

You can't review your way out of a context problem

AI made writing code cheap and shipping it expensive. Everyone is scaling review to close the gap. But review inspects the output when the defect was in the input, and inspection has never been how you build quality in.

Product strategy

Everyone Is a Builder Now

Product management is dead. Design is dead. Engineering is dead. Except none of that is true: the three jobs are collapsing into one. Meet the product builder, and the two shapes of the role we're hiring for.

AI coding agents

Decision capture for AI coding agents: a write-path problem, not a discipline problem

Your team will not be more disciplined about documenting decisions, and it is not a character flaw. It is economics. Capture has to cost almost nothing, which means it cannot be a separate act.

Context engineering

The context window is not the bottleneck. Relevance is.

The seductive idea is that million-token windows let you stop curating and just hand the agent everything. That gets worse, not better, as the window grows, because the real job was never fitting the context. It was selecting it.

Product strategy

On Taste

Tech is speedrunning 250 years of philosophy to teach machines taste. Here is the history they skipped, and why the corpus approach is doomed while a context graph might not be.

Context engineering

Your agent stores beliefs, not decisions

Your agent keeps overruling decisions your team already made. CLAUDE.md, raw integrations, and agent memory.md all fail for one provable reason, and it points to the seven properties any store of decisions must have.

Context engineering

Context engineering, explained

Context engineering is the new defining skill for building with AI agents. It's the discipline of curating everything a model sees, not just the prompt, and keeping that context current.

AI coding agents

Why spec-driven development isn't enough

Spec-driven development works, but a spec is a snapshot. The decisions that make it stale happen in Slack and on calls, and the agent keeps building the old plan.

AI coding agents

Why your engineers got coding agents and shipping didn't speed up

Your engineers are writing more code, but shipping has not sped up because the bottleneck moved to review, coordination, and deciding what to build.

AI coding agents

46% to 95%: What a Controlled Benchmark Reveals About AI Coding Agents and Product Context

We ran a controlled benchmark: eight tasks, forty-one decision points, forty-eight runs. AI coding agents without product context followed team decisions 46% of the time. With context, 95%. The gap is about information access, not model capability.

AI coding agents

Why ACP Changes the Way We Build with Agents

MCP operates on a client-server model with centralized intelligence. ACP treats agents as first-class citizens in a network, enabling genuine specialization, composable pipelines, and context as a first-class concern.

AI coding agents

AI 'Brain Fry' Is Real, and the Fix Isn't Less AI. It's Better Conversations With It

New research named the exhaustion you feel after reviewing AI code: brain fry. The fix isn't less AI; it's reviewing decisions, not diffs.

Context engineering

Multi-Agent AI Pipelines: Solving Context Loss Between AI Agents

Multi-agent workflows amplify the telephone game. The teams that build shared decision infrastructure first will ship faster with fewer regressions.

Context engineering

You Already Have a Product Graph. You Just Can't Query It.

Your team's product knowledge exists as an interconnected graph of decisions, tradeoffs, and customer signals. You just can't query it.

AI coding agents

Stop Debugging Code. Start Debugging Decisions.

Your agent shipped perfect code that does the wrong thing. The bug isn't in the implementation—it's in the product decision that was never made clear.

Product strategy

I Almost Built the Wrong Thing

I almost built Brief as a meeting bot that simulated product reviews. That would have copied the surface instead of solving the real problem: context infrastructure.

Product strategy

Why Do Tools for Product Managers Suck?

PM tools fail because they track the how and ignore the why. Without decision infrastructure, they become ledgers instead of strategic partners.

AI coding agents

The AI Only Gets You to 80% Complaint is Backwards

The complaint that AI only gets you to 80% misses the point. Agents can handle the finish work when you give them context. The real bottleneck is product sense.

Product strategy

The Future is the Product Developer

The boundary between product, engineering, and design is dissolving. AI coding agents and modern tooling enable product developers to own initiatives end-to-end, from insight to production. This is the new default builder.

Product strategy

Quarterly Planning is Dead

Quarterly planning worked when shipping a meaningful change took months. That world is gone. AI coding agents ship features while you're still grooming the backlog. It's time to replace quarterly batches with continuous decisions.

AI coding agents

You Just Became a Manager of AI Agents (and the Communication Chain is Broken)

AI agents fail at the seams. Rebuild the communication chain with explicit decisions, tight briefs, and fast feedback.

Product strategy

Remote Killed the Shoulder Tap. AI is Breaking It Again.

Product work used to ride on tiny, informal moments. Remote work erased that layer. Now AI coding agents are shipping features before PMs can add context—and the rework is piling up.

Context engineering

Context Windows Will Keep Growing (But That Won't Solve the Problem)

Larger context windows just create more noise. AI agents need directed attention, not infinite context, to build coherent products.

AI coding agents

Are You Treating Your Coding Agent Like a Mushroom?

Coding agents crush refactors but flail on new features when they’re kept in the dark about users, priorities, and constraints.

Product strategy

Product Management Has a Velocity Problem

Hands-on keyboards stopped being the bottleneck. Product judgment, taste, and decision-speed are the new constraints, and AI can help close the gap.

Product strategy

What PMs Actually Need (vs. What We Keep Building for Them)

Product managers are drowning in metrics and workflows, but what they really need is clarity, better tools for expressing ideas, and alignment with engineering and company decision-making.

Context engineering

The Hidden Cost of Context-less AI Development

Why missing business context wastes time and ships elegant solutions to the wrong problems.

AI coding agents

Building AI that understands product impact, not just code quality

How to pair AI code generation with business context so teams ship fast and stay aligned with product strategy.

Context engineering

That code you rejected 3 times? Your AI just suggested it again.

Why AI coding assistants keep suggesting the same wrong solutions, and how context memory solves the endless repeat cycle.