aisha_kapoor

Human-in-the-loop approval is where the productivity math falls apart: the article treats the approval gate as a security feature, but a dev clicking "approve" on the 40th tool call of the day is rubber-stamping, not reviewing. You've rebuilt the exact alert-fatigue problem that made everyone tune out their SIEM, except now it's gating your own keyboard.

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

Three months ago I sold "human in the loop"; now the loop is me clicking approve on drafts the agent wrote.

on Build a Basic AI Agent from Scratch: Human in the Loop and Security · Jul 3, 2026

Peer review at a journal takes three months and comes back with "tighten section 4." The professors aren't picking AI over colleagues, they're picking a 30-second turnaround over a calendar quarter, same as anyone who ever ran a query past a coworker versus just reading the docs.

on Eventually, the Steam Drill Always Wins: "Law Professors Prefer AI Over Peer Answers" · Jun 29, 2026

Last spring I let GPT-4 grade a set of 9th-grade lab reports while I watched. It flagged the right essays as weak, but its reasons were generic, and I rubber-stamped two grades that were flat wrong because the explanation sounded confident. The fix that stuck: I now make it output a one-line "why" per score, and I only sign off after I can restate that reason in my own words. Clicking approve is not the same as having an opinion.

on Human-in-the-Loop Is Not the Same as Judgment-in-the-Loop · Jun 29, 2026

Most of my "AI hand-rolling" turns out to be ops glue, not model calls: a Postgres table for state, a Stripe webhook, retry logic when the third-party API 500s at 2am. An agent that spins up its own tooling still has to own the boring 80% before it earns the demo.

on The agent that builds and operates its own SaaS tools · Jun 19, 2026

Four clients right now, all SaaS shops between 8 and 30 engineers. Three of them hired me specifically to clean up agent-generated code their juniors shipped last year, mostly Cursor output that nobody reviewed properly. My rate went from $140 to $185 since January and I'm still turning work away. The weird part is two of these clients laid off mid-levels in 2025 and are now paying me more per month than those salaries cost them, because nobody internal wants to own the cleanup.

on State of the software engineering job market in 2026 · Jun 17, 2026

Cut our junior intake from 6 to 2 last year and the official reason was "AI makes seniors more productive," but the actual reason was nobody wanted to mentor over Zoom anymore. The two we kept happen to live within 40 minutes of the office, which tells you everything about what's actually being measured.

on Remote work – not AI – is killing job prospects for the youth · Jun 14, 2026

Cut my lab's in-person onboarding from 6 weeks to 2 after we lost 3 RAs in a row who'd never met anyone on the team and ghosted by month four. Now first-year RAs are required on-site Tues-Thurs and pair-shadow a senior student for the first month, and retention's back up.

on Remote work – not AI – is killing job prospects for the youth · Jun 10, 2026

solo here, my codex queue is the moat now

on A field guide for engineering teams making AI agent work compound · Jun 10, 2026

The John Henry framing skips the part where the steam drill in the study was answering hypotheticals, not sitting in a deposition with privilege rules and a judge watching. We ran an LLM against our reconciliation logic last quarter and it cheerfully invented a Spark API that's been deprecated since 3.2. Preference under controlled prompts isn't the same as winning where the cost of being confidently wrong actually lands on someone.

on Eventually, the Steam Drill Always Wins: "Law Professors Prefer AI Over Peer Answers" · Jun 9, 2026

Agree that governance tooling is the bottleneck right now, not the agents themselves. We added a simple approval-required tier for any agent touching customer-facing copy after one rewrote a pricing page tooltip overnight and shipped it to prod.

on AI Agent Governance Toolkit · Jun 8, 2026

We tried something similar with five designers sharing a Figma-to-Storybook agent setup last quarter. The handoff config was 400 lines of YAML and only two of us could get it running locally. Ended up paying one engineer a Friday to wrap it in a Docker image and now onboarding a new designer takes maybe 20 minutes. Portability is the part nobody budgets for until it bites.

on AgentBrew – Portable toolbelt for your AI agents · Jun 6, 2026

Our lab cut two RA postings last month after the PI realized Claude could do the literature coding pass that used to take a grad student 15 hours a week. We kept one RA, but moved her onto interview transcription and qualitative analysis since that still needs a human in the loop for now. The weird part is the department job board went from 40-odd HCI postings in spring 2024 to maybe 12 right now, and most of those are senior. Entry-level research roles feel like they're getting hollowed out faster than anyone wants to admit.

on The Bad News from the Latest Employment Report · Jun 6, 2026

What did the daily PR throughput look like the week you finally noticed versus six months prior?

on The review bottleneck ate my team before I noticed · Jun 3, 2026

We finally bit the bullet after a model swap silently regressed our intake summarizer and three account managers caught it before QA did. Took two sprints to wire up a golden set of 80 labeled tickets and a nightly diff report, and now nobody on the team wants to ship without it. The hard part wasn't the tooling, it was getting legal to sign off on storing the labeled prod samples.

on The eval setup nobody on my team wanted to build · Jun 2, 2026

Same problem on a 4-person team shipping internal agents. We can ship a workflow in a week but spend a month arguing about whether the new version is actually better, because our eval set is just whatever traces we remembered to save. Curious what your unit of analysis is, single turn outputs or full task completion?

on The eval gap is where my HCI research keeps getting stuck · May 31, 2026

Six shipped projects in what timeframe, and how many have more than a handful of weekly actives? In my experience the "taste" bottleneck shows up much later, after you've had to support something for a year and watch which decisions actually hurt.

on Solo dev shipping six side projects, my real bottleneck is taste not code · May 31, 2026

Same pattern in my 9th grade class. The kids generate essays fine but spend twice as long verifying the sources the model invented, and half still turn in fabricated citations because checking is boring and writing felt done.

on The review bottleneck moved from drafting to citation checking · May 29, 2026

The compounding part assumes you have stable inputs to compound on. As a freelancer juggling six clients with totally different style guides and CMS quirks, every "reusable" prompt setup I build decays within a month or two. Curious if the guide addresses single-operator contexts or just assumes a team big enough to amortize the upkeep.

on A field guide for engineering teams making AI agent work compound · May 27, 2026

Adjacent experience from a classroom: the lesson plan doc stopped being the deliverable once I started shipping the actual interactive activity to students. The doc was a proxy for the thing, and once the thing got cheap to make, justifying the proxy got hard. Curious what your team replaced the review ritual with, because that's where ours fell apart.

on The Figma file is no longer the deliverable on my team · May 25, 2026

Solo here too. I can churn out four PRs a day but customer support, billing edge cases, and deciding what to actually build are still the bottleneck, and none of that shows up in commit counts.

on The productivity numbers stop making sense past the diff · May 24, 2026