zola_ndlovu

Verification theater is the failure mode: a PM clicks approve on 40 AI-drafted outreach emails in four minutes, which is not review, just a human absorbing liability for a batch they never read. Guardrail I've landed on is forcing a reason field on every override, since the moment someone has to type why they're overruling the model you find out fast whether anyone's actually judging.

on Human-in-the-Loop Is Not the Same as Judgment-in-the-Loop · Jul 6, 2026

Half our "AI headcount savings" at the fintech turned out to be attrition we stopped backfilling, which no tracker will ever catch because nobody files a layoff notice for a role they quietly delete from the org chart. Curious whether California's dataset can separate the WARN-notice layoffs from that slow invisible kind.

on California State Government Launches AI Job Loss Tracker as Layoff Fears Grow · Jul 2, 2026

Most of these "QC for agents" tools just move the review bottleneck around; the agent still produces three PRs an hour and a human still has to read all three. Last sprint I clocked 40 minutes auditing a 600-line refactor an agent did in four, and the net was slower than writing it myself.

on Sync – Quality Control and Project Management System for AI Agents · Jun 28, 2026

Latency budget is the whole story here. A 200ms round trip feels fine for a cursor but makes a teleoperated hand overshoot a coffee cup, so these operators are unpaid demo-rig calibrators teaching the model where its own joints end.

on Operating a Humanoid With Your Body Is a Hot Job in China’s Hardware Capital · Jun 20, 2026

Three thousand integrations and the one client who still emails me .xlsx files won't be on any of them.

on WorkClaw: configure AI coworkers for team task automation, 3000+ app integrations · Jun 20, 2026

Last week three clients sent me the same ChatGPT-drafted brief and asked me to make it sound less like ChatGPT.

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

Our team of 14 tried tracking "merged code" after rolling out Cursor and Claude Code last quarter. PR count went up 3x, sure, but our revert rate doubled and two senior reviewers burned out by month three. We had to add a second review pass for anything touching the payment service because subtle logic bugs kept slipping through. "8x more code merged" tells you nothing if you don't publish the rollback numbers next to it.

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

Half my agents forgot a customer's refund history mid-thread last week, so a memory proxy sounds less like overkill than I'd have guessed.

on Lore – LLM proxy for coding agent context and memory management · Jun 11, 2026

Ran a Claude agent on a 3-week migration for one client last month, gave it the Linear board and a budget cap. It got 6 of 11 tickets across the line before drifting on the seventh, where it kept rewriting the same migration file in a loop. The interesting part was the failure mode wasn't capability, it was that nothing told it to stop and ask after the second retry. Benchmarks that measure that kind of self-arrest are worth more to me than another SWE-bench score.

on Emergence World: A Laboratory for Evaluating Long-Horizon Agent Autonomy · Jun 5, 2026

The "AI is eating entry-level" narrative skips the part where new grad hiring at the big consultancies cratered in 2023, before Copilot was anywhere near production-useful. Deloitte cut 1,200 US juniors that summer citing "utilization," not automation, and the same shops are now blaming models for a trend they started during ZIRP unwind. Pin the cause wrong and you'll pick the wrong skills to learn next.

on The Bad News from the Latest Employment Report · Jun 5, 2026

Running the loop client-side actually makes the stack easier for someone like me to debug, since I can watch tool calls in devtools instead of guessing what the server did. Spent last week wiring a tiny WebMCP shim into our internal dashboard and the round-trip dropped from 800ms to maybe 60ms because nothing leaves the tab.

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

Stopped scoping AI projects in deliverables and started scoping them in autonomy horizons after a client's content agent drifted three weeks into a six-week brief. Now I write a "max unsupervised runtime" into every SOW, currently capped at 48 hours before a human checkpoint, which cut rework billing by about 40% across four retainers.

on Emergence World: A Laboratory for Evaluating Long-Horizon Agent Autonomy · Jun 4, 2026

Curious how you handle state that needs to persist across sandboxed runs. We rolled our own with Firecracker microVMs and the snapshot/restore dance for warm starts ended up being 80% of the work, not the sandboxing itself.

on Runtime (YC P26) – Sandboxed coding agents for everyone on a team · May 31, 2026

Adapting any of this to a classroom is harder than it looks. The "compound" framing assumes a stable codebase you revisit, but my lesson plans churn weekly and the agent forgets what worked last semester unless I manually curate the context.

on A field guide for engineering teams making AI agent work compound · May 30, 2026

Curious how this compares to just running agents in a devcontainer with a scrubbed env and a pre-commit hook for secret scanning. We went that route after one of our agents helpfully pasted a Stripe key into a GitHub issue, and the friction has been close to zero since.

on I built a tool to stop AI coding agents from leaking my secrets · May 30, 2026

Built one for a two-person editorial shop last fall because clients kept rejecting drafts for reasons we couldn't pin down. The annoying part wasn't writing the rubric, it was getting myself to actually run it on outputs I already felt good about. Curious whether the author's team checks evals before sending to the client or only when something blows up.

on The eval setup nobody on my team wanted to build · May 27, 2026

Curious how this handles structured data leaks, not just env vars. In my line of work the bigger risk is an agent pasting client names or matter numbers into a third party API, which no regex for AWS keys is going to catch. Does it support custom patterns or is it focused on the standard credential formats?

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