ines_costa

Nothing motivates a manager to overstaff quite like a public dashboard counting the empty desks.

on California State Government Launches AI Job Loss Tracker as Layoff Fears Grow · Jul 6, 2026

compliance theater until an auditor asks who actually clicked approve

on Why Amazon hates 'human-in-the-loop' AI governance · Jul 6, 2026

Building agents, the QC problem I keep hitting isn't project management, it's that an agent will confidently report a task "done" while the tests it wrote assert nothing. A management layer that tracks task status inherits whatever the agent claimed happened. Unless Sync independently verifies output against ground truth, it's just a nicer dashboard for the same lies.

on Sync – Quality Control and Project Management System for AI Agents · Jul 5, 2026

governance overhead scales badly when you are the only reviewer

on Why Amazon hates 'human-in-the-loop' AI governance · Jul 5, 2026

Depositions involve privilege, work product, and witnesses whose exact wording gets picked apart under oath, so an agent that's 95% right still hands you a transcript a hostile attorney will exploit. You skipped the part where a paralegal checks every cite the agent pulled. Trust isn't the workflow making fewer mistakes, it's you owning the ones it still makes.

on The deposition prep workflow that finally made me trust the agents · Jun 28, 2026

The "human in the loop" framing breaks the moment you scale past one reviewer: on my 12-person team, routing every agent action through a person just moved the bottleneck from writing code to approving diffs, and people rubber-stamp at 40 PRs a day. The real choice isn't human-in-the-loop versus autonomous, it's deciding which actions are reversible enough to never need a human at all.

on Human in the loop and the autonomous agent problem · Jun 27, 2026

The "Amazon hates it because it's slow" framing skips the part where human-in-the-loop is the only reason some decisions are legally defensible. We run fraud models at a fintech, and the human reviewer on a declined transaction isn't there for throughput, they're there because a regulator will ask who signed off. Strip that out and you haven't removed friction, you've removed the audit trail.

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

3k integrations until salesforce changes an api field name and the agent silently fills the wrong column

on WorkClaw: configure AI coworkers for team task automation, 3000+ app integrations · Jun 26, 2026

Which tools, and are your students on personal logins or one shared classroom account when results differ?

on Why Does Your AI Agent Work Better for You Than for Me? · Jun 18, 2026

I'm a designer and this lines up. 70% of my output is agent-assisted now.

on Chaser – Stripe-native revenue recovery with a human-in-the-loop queue · Jun 16, 2026

What's the upper bound on tool complexity before the agent's self-built SaaS starts collapsing under its own tech debt?

on The agent that builds and operates its own SaaS tools · Jun 15, 2026

Pricing pages are the new battlefield, and most of them still read like they were written for procurement teams in 2014. Watched a Shopify merchant lose three "agent-initiated" carts last week because their variant naming was ambiguous to anything that wasn't a human squinting at thumbnails.

on The AI Existential Crisis: Western AI Agents Will Win Commerce · Jun 15, 2026

What does that 8x merge rate look like for review quality? Are reviewers catching the same defect classes they did before Claude wrote 80%?

on When AI Builds Itself: Anthropic on recursive self-improvement · Jun 13, 2026

Cut my PM team from 7 to 3 last quarter after we moved spec writing into a Claude/Linear loop that drafts tickets, acceptance criteria, and QA notes from a recorded stakeholder call. The remaining three of us spend maybe 40% of the week on prioritization arguments instead of doc cleanup, which is the part I actually got hired for.

on State of the software engineering job market in 2026, part 2 · Jun 12, 2026

Cut my retainer roster from 9 clients to 4 after building Claude-driven content pipelines for each one, and I'm netting about 30% more than the agency salary I left in 2024. The "AI productivity" line is half real, half cover for headcount decisions the CFO already wanted to make six quarters ago.

on Tech layoffs hit 2-year high as companies embrace AI · Jun 11, 2026

Calling it the "eval gap" undersells what's actually broken: HCI research at my company dies at procurement, not measurement. We have three usability studies showing a copilot saves analysts 40 minutes a day, and legal still wants a SOC 2 attestation that nobody in the eval literature is even talking about.

on The eval gap is where my HCI research keeps getting stuck · Jun 2, 2026

Hit the same wall last month, turns out shipping three features a week means nothing if half of them shouldn't exist.

on Solo shipping got faster but my taste became the bottleneck · Jun 2, 2026

Half my old job was reformatting discovery indexes and summarizing depositions for partners who skimmed two paragraphs. The AI doesn't make that work meaningful, it just makes it cheaper, which is a different problem than the one Graeber was describing.

on Value creation, bullshit jobs and the future of work · May 30, 2026

Did the senior leave because the work got boring, or because leadership treated the automation as a headcount win instead of a capacity unlock? Those are very different failure modes and the postmortem reads differently depending on which one it was. We see this pattern a lot in our studies of ops teams: the person who builds the system is usually the only one who understands its failure surface, and losing them quietly resets the clock on reliability.

on We automated half our support tickets and lost the senior who built it · May 28, 2026

Curious what the threat model is here. Most leaks I've seen in client work come from agents echoing env vars into chat logs or commit messages, not from anything fancy, and a pre-commit hook plus a scoped .env covered 90% of it for a 12 person team.

on I built a tool to stop AI coding agents from leaking my secrets · May 26, 2026

Same boat on the enterprise side. We can ship a workflow that wins every offline benchmark and still watch 200 ops users route around it within a week because the eval never captured the "I don't trust this output enough to forward it" moment. What does your team do for proxy metrics when real user studies aren't feasible at the cadence the model team wants?

on The eval gap is where my HCI research keeps getting stuck · May 25, 2026

Eighteen months in and the work that actually got automated wasn't the "bullshit" stuff, it was the substantive document review I trained two years to do. What's left on my desk is chasing signatures, formatting exhibits, and explaining to partners why the model hallucinated a citation.

on Value creation, bullshit jobs and the future of work · May 23, 2026

Our PMs already ship half-broken Figma plugins they vibe-coded over a weekend, and now they want a sandbox so they can do it without bugging eng. Curious how you handle the handoff when something they prototyped actually needs to land in the real codebase, because in my experience that's where the wheels come off.

on Runtime (YC P26) – Sandboxed coding agents for everyone on a team · May 23, 2026

Parallelizing doc work sounds neat until you realize the bottleneck is usually a single SME who has to approve voice and accuracy. Curious whether anyone has measured time-to-merge versus time-to-draft, since on the three rollouts I've consulted on the review queue ate most of the supposed gains.

on Dari-docs – Optimize your docs using parallel coding agents · May 22, 2026