780 posts tagged with "ai-engineering"

A user reports the assistant gave them a wrong answer at 9:47 a.m. You open the trace. There are three hundred and forty spans. They are almost all named agent.run, llm.invoke, or tool.call. Some have a parent. Some are siblings. Three of them retried. One of them retried and then was cancelled. None of them tells you whether the bad output came from the planner, the worker, the critic, the reflection pass, or the second retry of the worker after the critic flagged it.

You spend the next hour grepping log lines for a UUID prefix you saw in a screenshot, cross-referencing timestamps against a Slack notification, and reconstructing the agent topology in your head from the indentation pattern in the trace viewer. Eventually you guess that the third worker invocation ran with a model alias that silently flipped to a different snapshot the night before. You cannot prove it from the trace alone.

The agent worked. The trace is intact. The hairball is the bug.

A staff engineer I worked with last quarter caught a bug that had already been "fixed" three times in the previous six weeks. Three different engineers. Three different files. Three green CI runs. Three accepted agent-generated patches. Each patch made the failing test pass and the user-reported error disappear. Each one moved the bug somewhere else, where it waited until a different surface area triggered it again. The fourth time it surfaced, the data corruption it caused had been silently compounding for forty days.

The bug was a single off-by-one in a pagination cursor. The agent had been right that the symptom would go away. It had been wrong about why. And the engineers — competent, senior, well-intentioned — had each accepted a passing patch before they understood the failure mechanism.

This is the agentic debugger's trap: your agent can produce a fix faster than you can build the mental model needed to evaluate whether the fix is correct. Patch velocity outruns diagnosis. The bug count drops, the CI dashboard goes green, and you ship a codebase whose failure modes you no longer understand.

In 1964, thirty-eight people watched Kitty Genovese being attacked outside their apartment building in Queens. None of them called the police until it was too late. Latané and Darley spent the next decade explaining why: the more people who can see a problem, the less likely any single one of them is to act. They called it diffusion of responsibility. In their famous seizure experiment, 85% of participants intervened when they thought they were alone with the victim. When they believed four others could also hear the seizure, only 31% did.

Now picture your last AI feature launch. Product wrote the prompt. Engineering picked the model and wired the gateway. The data team curated the retrieval corpus. Safety bolted on the input and output filters. Customer support drafted the escalation playbook. Five teams in the room. Each one shipped its piece on time. Three months in, the feature's accuracy has quietly slid from 89% to 71%, the eval suite has not been run since launch week, and when you ask who owns the regression, every team can name three other teams that own it more.

Picture the scene: it is 2:14 a.m., the on-call engineer's phone is buzzing, and the AI feature that ships your flagship product surface is confidently telling enterprise customers that their account number is "tomato soup." The model provider pushed a routing change, your prompt got truncated by a quietly upgraded tokenizer, or the retrieval index regenerated against a corrupted parquet file — the cause does not matter yet. What matters is the ten-minute clock until someone screenshots an output and posts it to LinkedIn.

If your only response is "revert the deploy and wait for CI," you have already lost. A standard pipeline rollback is twenty to forty minutes from page to recovery, and the bad outputs do not pause politely while the green checkmark renders. By the time the new container is healthy, the screenshot is in a thread, the support inbox has fifty tickets, and the trust you spent six months building is being audited by people who never use the product.

The teams that contain these incidents in five minutes instead of five hours did not get lucky. They built a kill switch before they needed one — a primitive that lets the on-call engineer disable the AI path in seconds without a deploy, without a merge, and without anyone touching the production binary. This post is about what that primitive looks like for AI features specifically, why the deterministic-software version of it is insufficient, and what has to be true the day before the incident for the response to work the night of.

The two-week canary is one of those practices that sounds disciplined enough to skip the harder question. Engineering imported it from microservices — ramp 1% for a few days, watch error rate, ramp to 100%, declare done — and grafted it onto AI features without asking whether the failure modes that matter for AI even surface in two weeks. They don't. The bill that kills the feature lands in week six. The customer cohort that exposes the long-tail intent onboards in week five. The eval drift that scored +3% on launch day starts costing real money in week four because the new prompt's chattier outputs have been compounding token spend the whole time, and nobody was watching for that because the dashboard was watching for crashes.

A canary built around p95 latency and HTTP 500s will tell you the LLM is up. It will not tell you the feature is working. AI features fail in shapes the deploy ceremony was never designed to catch — slow shape changes in user behavior, gradual cache erosion, retrieval quality collapse, refusal-rate creep, cost trajectories that bend the wrong way — and almost all of them take longer than two weeks to declare themselves. The team that ships by the microservice clock is shipping by a clock the failures don't run on.

The classic bug bash is a deterministic ritual built for deterministic software. Ten engineers crowd a Slack channel for two hours, hammer a checklist of golden-path flows, and file tickets with crisp repro steps: "Click X, see Y, expected Z." It works because the system under test is reproducible — same input, same output, same bug, every time.

Run that exact ritual against an AI feature and you will produce two hundred tickets, close one hundred and eighty as "expected stochastic variation," and miss the twenty that signal a real cohort regression. The format isn't just stale; it's actively miscalibrated. A bug bash against an LLM-backed feature is not a defect-hunting session. It is a sampling exercise against a probability distribution, and the team that runs it like a deterministic test session is collecting noise and calling it signal.

This post is about how to redesign the bug bash for stochastic systems — what to change about the format, the participants, the triage rubric, and what counts as "done."

A multi-agent system for market data research quietly burned through $47,000 in inference cost over four weeks before anyone noticed. The original weekly bill was $127. The cause wasn't a traffic spike or a model upgrade — it was two agents passing the same conversation back and forth for eleven days, each one confident the other was the right place for the request to live. Nothing errored. No alarm fired. The bot's "queue transferred" metric and the other bot's "task received" metric both went up in lockstep, and both dashboards looked healthy.

This is the closed-loop escalation bug. It is the multi-agent version of two helpful colleagues each insisting "no, you take it," except neither of them ever gets bored and walks away. The architecture diagram you drew at design time has each specialist owning a clean slice of the problem. The architecture the runtime actually executes has a routing cycle nobody in the room can see.

Every AI feature ships with a moment its team hasn't planned for: the moment it has to be turned off. A model regression lands during the morning standup. A cost spike from a marketing campaign nobody told engineering about doubles the bill in twelve hours. A privacy review flags a prompt-context leak. The provider goes down for ninety minutes. A compliance team waves a flag at noon and the feature has to disappear before the close of business.

The disable switch most teams ship for that moment is "the feature returns an error" — a spinner that never resolves, a banner that says "AI assistant unavailable, try again later." That is a strictly worse user experience than the pre-AI status quo, which is exactly what users will compare you to the moment AI degrades. The status quo had a button. Now they get an apology.

A frontier-model chat feature is roughly a thirty-cents-per-conversation product. The distilled variant of the same feature is roughly a third-of-a-cent-per-conversation product. These are not two implementations of one product. They are two products, with different free-tier economics, different acquisition costs, different markets, and different competitive moats. The team that ships the distilled version as "the same feature, cheaper" wastes the move.

Most engineering organizations still treat distillation as a research-team optimization that gets applied after a feature is "done" — a tail-end pass to wring inference cost out of something already spec'd against the frontier model. That framing is wrong by an order of magnitude. The choice of teacher, the choice of student, the eval suite the student is graded against, and the product surface the student is deployed to are product decisions. They determine which capabilities you are consenting to lose, which traffic shape you are designing for, and which price floor you are unlocking. Hand them to a research team to optimize against MMLU and you will ship a model that wins benchmarks the product does not care about.

The number that decides whether a model goes to production is being produced by a Jupyter notebook running on a single engineer's MacBook, against a CSV that lives in a Slack DM, scored by a judge model that nobody pinned. Two weeks later, after the engineer has touched the notebook three more times and the API provider has silently shipped a minor model update, nobody on the team can reproduce the number — including the engineer who originally generated it. And yet that number is the gate. It decided that GPT-4o-mini was good enough to replace GPT-4 in the customer support flow. It decided the new prompt template shipped. It decided the fine-tune was promoted. The team is treating it like a load-bearing artifact and storing it like a sticky note.

This is the eval gap. The industry has spent five years writing about evaluation as a methodology problem — which scoring technique, which judge model, which rubric, which dataset — and almost no time writing about evaluation as an engineering problem. But the moment your eval suite starts gating production releases, it inherits every requirement that the rest of your production stack lives by: reproducibility, version control, ownership, observability, dependency management, latency and reliability budgets, and a pipeline that survives the engineer who built it leaving the team. Most teams skip this layer entirely and discover its absence only after a major incident, usually one where the eval score said green and the customer experience said red.

Your eval pipeline scores are trending up. Response quality is improving. The LLM judge is catching more bad outputs. Your dashboard is green.

Meanwhile, a support ticket trickles in: "The assistant keeps giving me long, formal answers when I asked a simple question." Then another: "It stopped suggesting next steps. Used to do that automatically." Then your product manager shows you a chart: user satisfaction down 12% over the last quarter, correlated almost perfectly with the stretch where your automated eval metrics were climbing fastest.

This is the eval automation trap. Your measurement apparatus became optimized for itself rather than for what your users value — and because the feedback loop was entirely automated, nobody noticed until the damage was already in production.

Every AI team I've watched ship a "small" output schema change has lived through the same week. Someone renames a field in the system prompt — say, summary becomes tldr, or the tool catalog gains a required confidence parameter — and the next CI run lights up red across 800 eval cases that have nothing to do with the change. The prompt diff is fifteen lines. The eval diff is a four-day migration project nobody scoped, owned, or budgeted.

This is the eval migration tax. It is the maintenance cost no roadmap accounts for, paid in delayed releases that get blamed on "flaky tests" rather than the architectural choice that actually caused them. Most teams pay it for years before they recognize the pattern, because each individual incident looks like ordinary churn. The compounding only becomes visible when you tally the engineering hours spent migrating evals across a quarter and realize they exceed the hours spent improving the model behavior the evals were supposed to measure.