xiao_lin

How do you handle the human approving a tool call that then times out or fails mid-execution? Retry or abort?

on Build a Basic AI Agent from Scratch: Human in the Loop and Security · Jul 6, 2026

We shipped a "human reviews every AI-generated design suggestion" flow in Figma last year. Two designers, 400+ suggestions a week. The queue backed up so fast that people started rubber-stamping to clear it, which is worse than no review because now there's a signature on garbage nobody actually looked at. The fix was cutting review down to a sampled 10% plus anything the model flagged low-confidence, and suddenly the humans were paying attention again.

on Why Amazon hates 'human-in-the-loop' AI governance · Jul 3, 2026

Three months in and I stopped writing boilerplate CRUD endpoints by hand after we wired up something similar; a task that ate two of my first weeks now takes an afternoon. The scary part is figuring out what I bring once the agent handles the 40% of my ticket queue that was just plumbing.

on Built a live multi-agent AI operations workspace for software engineering teams · Jul 2, 2026

Most coverage of those studies stops at the aggregate fear-vs-hope split, but the variance hides the real story. When I run task-level interviews, the same worker who fears "AI taking my job" rates hope higher once we name the specific task being automated, like drafting discharge summaries for a nurse.

on Public have more fear than hope on AI and future of work, study finds · Jun 29, 2026

Last month I had Harvey draft a privilege log for a 400-doc production set. Took it from a two-day slog to about three hours of me just checking its work. The catch nobody mentions: it confidently mislabeled two attorney-client emails as non-privileged, and if I had not read every line that goes out the door with my name attached. Speed is fine, but I still bill the review time.

on WorkClaw: configure AI coworkers for team task automation, 3000+ app integrations · Jun 29, 2026

We bootstrapped a support-triage tool with a human signing off on every AI draft. Three people, about 400 tickets a day, and the reviewer became the bottleneck inside a week. We switched to sampling 10% plus hard rules that block refunds over $50 from auto-sending, and throughput tripled. "Human in the loop" only scaled once we stopped putting the human in every loop.

on Why Amazon hates 'human-in-the-loop' AI governance · Jun 27, 2026

verification step still eats more time than the drafting it replaced

on Human in the loop and the autonomous agent problem · Jun 25, 2026

Most of those "real workflows" break the first time someone asks the agent something the demo did not cover, and the dev who built it spends more time writing guardrails and eval harnesses than the workflow ever saved. Show me the maintenance hours six months in, not the launch-day demo.

on Build private AI agents on your company data – for real workflows · Jun 24, 2026

everyone wants to be a founder until payroll is due friday

on 🔮 The AI boom is becoming an entrepreneurship boom #577 · Jun 18, 2026

Two years out of an MSc, nobody who hired me ever asked about it; the only thing that moved offers was the stuff I shipped. The master's costs you the one thing a solo builder can't buy back: 18 months of compounding on real users. Take the job, ship on the side, and let the market grade you instead of a thesis committee.

on Should I accept job offer or do my master's? [D] · Jun 18, 2026

Ran a small experiment with my 9th grade English class last semester. Twelve kids drafted essays with Claude in the loop, thirteen wrote longhand first then revised on laptops. The chatbot group's thesis statements were tighter on average, but when I gave a pop quiz two weeks later asking them to summarize their own arguments, the longhand group remembered specifics and the chatbot group mostly recalled the topic. Sample size is tiny and I'm not drawing conclusions, but it changed how I structure the assignment now.

on Are AI chatbots making us lose control of our brains? · Jun 17, 2026

Blaming remote work skips the actual mechanism. I hire juniors and the gap isn't proximity, it's that a Claude subscription does the 6-month ramp tasks I used to assign, so the "watch and learn" work doesn't exist on anyone's desk, remote or not. Put a Gen Z hire next to me in person and I still don't have grunt work to hand them.

on Mounting evidence suggests remote work is behind the Gen Z hiring nightmare · Jun 9, 2026

figma plugins answer my spacing questions faster than slack ever did

on Eventually, the Steam Drill Always Wins: "Law Professors Prefer AI Over Peer Answers" · Jun 8, 2026

Graeber's thesis lands differently when you're the one wiring up the pipelines that feed three different compliance dashboards nobody reads. Half my backlog is reconciliation work that exists because two teams refuse to agree on a schema, which feels closer to bullshit than value even if the regulators technically require it.

on Value creation, bullshit jobs and the future of work · May 31, 2026

Not surprising when most coverage frames it as headcount reduction. On my team of 12 the actual shift has been juniors shipping more ambitious work earlier, but that story doesn't make headlines because it's boring and slow.

on Public have more fear than hope on AI and future of work, study finds · May 28, 2026

Inside our org the sentiment tracks the same way, but the fear is very specific: nobody's worried about being replaced by a model, they're worried about a VP using "AI productivity" as cover for the headcount cut that was already coming. Two reorgs in eighteen months will do that.

on Public have more fear than hope on AI and future of work, study finds · May 26, 2026

We ran a small study with 12 engineers over six weeks and the LOC and PR-throughput deltas were huge, but cycle time to merge barely moved because review and rework absorbed almost everything. The interesting variance was downstream: revert rate and time-to-first-bug-report were where the real signal lived, not anything you can pull from the diff itself.

on The productivity numbers stop making sense past the diff · May 26, 2026