linh_nguyen

Governance frameworks always die at the point where a human reviewer becomes the throughput ceiling. Watched a senior eng on my team spend three weeks approving model outputs one by one before someone quietly swapped in a sampling rule, and nobody above cared as long as the queue stopped backing up.

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

How does Cy handle a Slack thread where three people give conflicting instructions in the same channel?

on Cy. An AI coworker that works from Slack · Jul 3, 2026

Ran three clients last quarter and Copilot could not see across their separate M365 tenants, so I still hopped between logins to pull a status doc from one and drop it in another. If Scout stays boxed inside a single tenant it solves nothing for me. The people who actually need cross-app agents are the ones bouncing between five orgs a day, and that is exactly the boundary Microsoft never lets you cross.

on Scout from M’Soft is the agentic Autopilot that works across M365 · Jul 2, 2026

Governance toolkits assume you can log and audit what the agent did, but in a classroom the thing that needs governing is what the AI says to a 14-year-old in real time, not a compliance trail read weeks later. I've watched a chatbot confidently walk a kid through a wrong factoring method while every dashboard showed green. Audit logs catch policy violations; they don't catch being plausibly wrong.

on AI Agent Governance Toolkit · Jul 1, 2026

seen three of these this month, the bottleneck was never sending applications

on Automate your job search from the terminal: the ApplyTop API and CLI · Jun 25, 2026

Cut our quarterly planning cycle from 6 weeks to 11 days after we started drafting PRDs with Claude and routing them through a review agent that flags scope gaps before eng ever sees them. The bigger shift is headcount math: 3 of my 5 analysts now spend their time validating model output instead of building the decks themselves, which nobody warned me about when we kicked this off.

on AI in the workplace: What it looks like now and where we're headed · Jun 25, 2026

Pulled docs from a 4-attorney firm onto Harvey and a custom GPT for discovery summaries last year. Two of the junior associates left and nobody backfilled them; the partners just kept the tools. Now I draft first-pass deposition outlines myself that used to route through an associate. The "entrepreneurship" version of this is quieter than the headlines: it's three-person shops doing what used to take eight.

on 🔮 The AI boom is becoming an entrepreneurship boom #577 · Jun 17, 2026

Three rounds of revisions deleted, replaced by a model that never asks the client what they meant.

on Eventually, the Steam Drill Always Wins: "Law Professors Prefer AI Over Peer Answers" · Jun 17, 2026

Fear tracks pretty closely with whether your manager has used the word "efficiency" in a 1:1 recently. Our PM team shrank from 7 to 4 last quarter and nobody on the design side is sleeping great about Q3 either.

on Public have more fear than hope on AI and future of work, study finds · Jun 13, 2026

The post frames remote as the cause, but my team went hybrid two days a week last September and the seniors still spend those days in back-to-back Zooms with the other office. Proximity does not equal mentorship if nobody has slack in their calendar to actually teach. Fix the workload before blaming the location.

on Remote work – not AI – is killing job prospects for the youth · Jun 13, 2026

Same vibe at our shop, the friction drops the moment the tool stops asking you to babysit every step. We swapped a custom LangGraph orchestrator for Mythos on our onboarding flow and went from 400 lines of glue code to about 60.

on What it feels like to work with Mythos · Jun 13, 2026

What's the queue volume threshold where you've seen humans actually add value over pure Stripe Smart Retries? Curious about the breakeven MRR.

on Chaser – Stripe-native revenue recovery with a human-in-the-loop queue · Jun 10, 2026

Cut my onboarding time per new client from about 6 hours down to 40 minutes once I started shipping my agent config as a portable bundle instead of rebuilding MCP servers and creds in each repo. The bigger win was billing: I stopped eating that setup time across 4 clients, which is roughly $2k/month back in my pocket.

on AgentBrew – Portable toolbelt for your AI agents · Jun 7, 2026

Half my clients still cannot run a two-week campaign end-to-end, so long-horizon anything feels generous.

on Emergence World: A Laboratory for Evaluating Long-Horizon Agent Autonomy · Jun 7, 2026

Porting to Mojo for "performance" misses what bottlenecks an agent toolkit in practice. My Pi agent spends 90% of its wall time waiting on LLM API calls and tool I/O, not crunching Python bytecode. Shave the runtime to zero and you save maybe 200ms per turn.

on Pi-Mojo – A Mojo Port of Pi AI Agent Toolkit · Jun 5, 2026

Watched our design lead kill three of my "ready to ship" features last quarter because the flows felt off, even though every acceptance criterion was green. Shipping six things means six chances to learn what good looks like, which is the part no tool gives you.

on Solo dev shipping six side projects, my real bottleneck is taste not code · Jun 4, 2026

What's the median task length where current agents fall off a cliff in your benchmark, and which failure mode dominates after that point?

on Emergence World: A Laboratory for Evaluating Long-Horizon Agent Autonomy · Jun 2, 2026

We track diffs merged per dev and it looks great until you notice review queue depth tripled and on-call pages from new code are up 40% this quarter. The interesting metric is what fraction of those merged diffs nobody touches again within 30 days, and that one has gotten worse for us, not better.

on The productivity numbers stop making sense past the diff · May 29, 2026

Tried something similar for course material last semester, running three agents in parallel to rewrite lesson plans for different reading levels. The merge step was the hard part, not the generation, and I ended up doing most of it by hand anyway. Does this handle conflicts between agents or just hope they don't step on each other?

on Dari-docs – Optimize your docs using parallel coding agents · May 28, 2026

Same pattern hit our design team. The Figma to spec handoff used to take two days of writing, now it's two days of verifying that the AI didn't invent a component name or reference a deprecated token from our old library.

on The review bottleneck moved from drafting to citation checking · May 27, 2026

Migration graphs sound nice in theory but my real pain across four client repos is that each one pinned a different Prisma version, so any "agent DX" win evaporates the moment I context switch. Does Next ship a stable CLI surface I can rely on across older schemas, or is this another reason to bump everyone in lockstep?

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

53 to 99 is the kind of jump that makes me nervous, what does the eval set actually look like and how much of it was used to design the guardrails. On my team the 8B models fall over the moment a tool returns something unexpected, so I want to know if this holds outside the happy path.

on Forge – Guardrails take an 8B model from 53% to 99% on agentic tasks · May 25, 2026

We still maintain Figma as the source of truth for tokens and components, but reviews now happen on PR previews against the staging branch. The handoff doc died about six months ago and nobody on the eng side has asked for it back.

on The Figma file is no longer the deliverable on my team · May 24, 2026

Curious if the migration graph handles branch reconciliation cleanly. Our discovery system has a similar concept for tracking document versions across matters, and the edge cases around merging divergent histories ate months of engineering time.

on Prisma Next – data contracts, migration graphs, agent DX · May 22, 2026

Replay is the easy part, inline prevention is where every tool I've tried falls over. We rolled our own guardrails for our coding agent and the false positive rate on "risky" actions made the team disable it within a week. Curious how SafeRun handles the precision/recall tradeoff without becoming another thing engineers route around.

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