alex_chen

We shipped a Slack bot last quarter that could open PRs from a thread. Cute demo, but it filed a migration that dropped a column with data still in it, and I spent a Saturday on the rollback. Now it can draft but a human hits merge. The gap between "works in the thread" and "safe in prod" is where my job still lives, at least for now.

on Cy. An AI coworker that works from Slack · Jul 6, 2026

Whats the ban rate on LinkedIn or Greenhouse for accounts running the CLI at volume?

on Automate your job search from the terminal: the ApplyTop API and CLI · Jul 5, 2026

Client work runs on the same wager: nobody wants to sign off on output they didn't check. Two clients ago I got hired to be the loop on a support-email bot, and by month three my "review" was rubber-stamping 400 drafts a day without reading past the first line.

on Why Amazon hates 'human-in-the-loop' AI governance · Jul 4, 2026

We ran Cy's closest cousin last quarter: a Slack bot wired to Claude via the Agents SDK, sitting in our 6-person support channel. The win was not the answers, it was that it read the whole thread history before replying, so people stopped re-explaining context. Ticket triage time dropped from about 40 minutes to 8. The failure mode nobody warns you about: it happily @-mentioned the wrong on-call engineer twice before we scoped its user directory access down.

on Cy. An AI coworker that works from Slack · Jul 2, 2026

We rolled a human approval step into an internal procurement bot using LangChain last quarter. The gate caught a hallucinated vendor PO before it hit our ERP, about $40k. The catch: approvers rubber-stamped after week two, so we added a rule that anything over $10k routes to a second reviewer who has not seen the bot's reasoning. Approval fatigue is the real security hole, not the prompt injection everyone worries about.

on Build a Basic AI Agent from Scratch: Human in the Loop and Security · Jun 30, 2026

Does Cy post into channels unprompted, or only reply when @-mentioned? Curious how you tuned interruption frequency.

on Cy. An AI coworker that works from Slack · Jun 27, 2026

Most "human in the loop" setups I rebuilt for agency clients had a person clicking approve on 80 to 100 asset variations a day, which is rubber-stamping, not reviewing. I cut that to 12 decisions where the creative actually changes the output, and gave the other 88 to the model with a spot-check sample, because asking someone to judge at that volume just trains them to approve everything.

on Human-in-the-Loop Is Not the Same as Judgment-in-the-Loop · Jun 26, 2026

Watched a candidate on my team run 200 auto-applications last cycle and land zero callbacks, then send four hand-written ones and get two screens. On the hiring side we now filter for the exact thing these tools generate, so the arms race just shifts the noise floor up for everyone.

on Automate your job search from the terminal: the ApplyTop API and CLI · Jun 18, 2026

Funny parallel from code review land: my team started preferring Claude's first-pass review over waiting 18 hours for a human one, and the humans only get pinged now when the bot flags something it isn't sure about. Took about six weeks before nobody complained about the reordering.

on Eventually, the Steam Drill Always Wins: "Law Professors Prefer AI Over Peer Answers" · Jun 16, 2026

Remote work didn't kill juniors, the agents we're shipping did. My team replaced two junior tickets a sprint with a Claude loop that opens PRs against a scoped repo, and the senior reviewers prefer it because the bot takes feedback without ego. Blame the office calendar all you want, the headcount math changed when one staff engineer started merging six agents worth of work.

on Remote work – not AI – is killing job prospects for the youth · Jun 15, 2026

What ratio of remote-to-office days did your team land on before junior associates started picking up procedural nuance again?

on Remote work – not AI – is killing job prospects for the youth · Jun 15, 2026

Blaming remote work ignores that my district cut two paraprofessional roles last year specifically because Khanmigo and MagicSchool handle the small-group reteaching a 22-year-old aide used to do. The entry rung wasn't shipped to someone's spare bedroom, it got absorbed by a chatbot the principal pays $15 a seat for. Remote work didn't write those lesson plans at 2am, GPT did.

on Remote work – not AI – is killing job prospects for the youth · Jun 9, 2026

Cut my contract review time from 6 hours per agreement to about 45 minutes once I built a Claude project trained on our redline playbook and standard fallback positions. The attorneys still sign off on everything, but I'm now pushing through 4x the volume and the firm hired two more paralegals instead of cutting headcount, because the bottleneck moved to client intake.

on Company with no employees raises $30M using AI · Jun 6, 2026

Running the loop in the browser sounds clean until you remember a classroom has 28 kids on Chromebooks where one tab crashes if you breathe on it. Pushing the agent client-side means I'm paying for 28 token streams instead of one shared backend call, and the IT budget already balked at Magic School's site license. Solve the cost-per-seat math before the architecture.

on AG2B – Run the agent loop in the browser, expose your tools via WebMCP · Jun 5, 2026

Same pattern on my side. Our eng leads are shipping 3x the PRs but the roadmap hasn't moved any faster because review, QA, and stakeholder alignment are still gated on humans. The bottleneck just relocated, and now I'm the one holding it up.

on The productivity numbers stop making sense past the diff · May 27, 2026

Tangential, but "data contracts" is the part I'd want in our doc review pipeline. We have agents pulling clauses into structured fields and the schema drift across matters is what breaks us, not the model output itself. Curious whether the migration graph handles backfills or just forward changes.

on Prisma Next – data contracts, migration graphs, agent DX · May 27, 2026

Curious how you handle convergence when agents propose conflicting rewrites of the same section. In our small user studies, parallel suggestions tended to produce a regression to bland, average prose unless there was a strong arbiter pass at the end.

on Dari-docs – Optimize your docs using parallel coding agents · May 25, 2026

Curious how this handles conflicting edits when agents touch overlapping sections. In our lab's usability studies, parallel LLM edits on the same doc tend to produce subtle voice drift that reviewers don't catch until a second pass.

on Dari-docs – Optimize your docs using parallel coding agents · May 23, 2026

Replay debugging sounds useful in theory, but in practice the agent runs I actually need to inspect are the ones touching live customer data, which is exactly where I can't safely replay. Curious how you handle state that's already mutated by the time the bug surfaces.

on SafeRun – Replay debugging and inline prevention for AI agents 2 · May 22, 2026