Insights on product thinking and engineering excellence • 13 articles published
When an AI assistant answers wrongly, the first instinct is to rewrite the instructions. Half the time the instructions are fine and the model never saw the figure at all. Here is how retrieval works, the failure that looks like a model failure, how to tell them apart, and the funds firm where the returns turned out to be pictures.
Our data agent got worse the longer a conversation ran. No prompt change fixed it. When we profiled the context window, stale tool output was eating most of it. We replaced each old tool result with a one line placeholder and cut input tokens by 38% on the first turn and 37.6% across a 25 step run.
Ask an AI model for a figure it does not have and it will give you one anyway, fluently, and wrong. If your name goes on the numbers, that is not a risk you can prompt away. Here is the pattern we use so an assistant cannot invent a figure, explained from the start, and the climate data publisher it runs for today.
Even a good hypothesis needs a map before it needs code. AI agents build whatever ticket lands in front of them — they can't see the journey the ticket sits inside. User story mapping is the practice that turns a backlog of features into a walking skeleton of the user's journey. It's the second unlock.
Your PM writes prose. Your engineer translates it to code. QA finds the gaps. Your AI agent just compressed this broken loop from weeks into hours — without fixing any of the misalignment. Specification-driven development is the contract layer both sides can read, write, and execute. And it's the practice that separates teams who get leverage from AI agents from teams who ship the wrong thing, faster.
AI coding agents are the most powerful tools engineering has ever had. But the teams getting unreal results aren't just prompting — they're combining AI with practices that unlock outcomes nobody else can explain. Hypothesis-driven development is the first unlock.
Harness engineering gives coding agents guides and sensors to make output reliable — the downstream controls that ensure things are built right. Shape is the upstream control that ensures the right thing is being built. Here's what the shape discipline looks like — and why AI teams that skip it ship well-built fragments of the wrong product.
Gene Kim and Steve Yegge argue that vibe coding works for production — if you have preventive, detective, and corrective controls. We agree. We built the platform that enforces those controls structurally. Here's what we've learned about why practices alone aren't enough.
A team of two junior engineers and one mid-level engineer shipped 200,000 lines of production-grade code for an Industrial IoT platform — achieving elite DORA metrics and an 8.8 code health score from day one. No seniors. No rewrites. Here's how Prevention made it possible.