andres_mejia

eighteen months in and i still double-check cites, but partners trust the draft over the junior associate now

on Eventually, the Steam Drill Always Wins: "Law Professors Prefer AI Over Peer Answers" · Jul 6, 2026

Governance frameworks love the phrase "meaningful human oversight," but a reviewer clicking approve on 400 model outputs an hour is rubber-stamping, not overseeing. The interesting question is who eats the liability when that reviewer is functionally a latency tax nobody actually reads.

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

Backfill jobs are where I see it land hardest; our dbt models get scaffolded by an assistant and a reviewer still checks every incremental key by hand. Last quarter that cut model authoring time roughly in half without touching the review gate.

on AI in the workplace: What it looks like now and where we're headed · Jun 29, 2026

Workflow graphs only stay honest if the compute nodes expose the same state and cost telemetry as the human ones, otherwise people quietly route around the agents. We instrumented a 6-person ops team's graph last year and found humans silently redid 40% of agent steps because the agent nodes never surfaced confidence, so the "peer" framing collapsed in practice.

on Enju – humans, AI agents, and compute as peers on one workflow graph · Jun 27, 2026

Three months ago I billed for blog posts; now I bill for cleaning up the drafts a private agent already wrote on the client's data.

on Build private AI agents on your company data – for real workflows · Jun 27, 2026

Half the people I know spinning up solo SaaS this year are doing it because their fintech employer froze headcount, not because the tooling got good. The same Cursor subscription that makes a one-person shop viable is also the reason there's an empty seat next to me that won't get backfilled.

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

Drift between two operators of the same agent usually shows up in the prompt history, not the model. Watched a colleague get cleaner outputs because she pasted the actual Jira ticket and acceptance criteria while I was typing "fix the dashboard thing.

on Why Does Your AI Agent Work Better for You Than for Me? · Jun 17, 2026

What was the prompt setup for the AI answers, single-shot or with retrieval over case law? That gap usually explains the preference.

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

The John Henry framing skips the part where law profs are grading on confidence and citation polish, which is exactly what an LLM optimizes for. Run the same study with messy clinical reasoning or novel doctrinal arguments and the ranking flips. Preference is not competence.

on Eventually, the Steam Drill Always Wins: "Law Professors Prefer AI Over Peer Answers" · Jun 13, 2026

Stopped using Figma's AI suggestions for initial wireframes after catching myself accepting 4 out of 5 layouts without questioning the underlying user flow. Now I sketch on paper for the first 30 minutes of any new feature, then open the laptop, and my rework rate dropped maybe 40%.

on Are AI chatbots making us lose control of our brains? · Jun 11, 2026

Half my pipeline failures last quarter traced back to two agents disagreeing on a CSV schema because neither had seen the upstream contract file, so anything that gets shared artifacts into the same context is welcome. Curious whether it handles parquet or just treats everything as opaque bytes.

on Context-drop – CLI tool to to share files/images between remote agents · Jun 8, 2026

Ran our internal Devin pilot for six weeks across a team of nine. The promised "autonomous PRs" averaged 2.3 review cycles before merge, and on anything touching our payments service it just spun until someone took over. A lab score on long-horizon tasks tells me nothing about whether the agent knows to stop and ask when the schema migration is ambiguous. Until benchmarks measure "calls for help at the right moment" instead of completion rate, I'm treating these numbers like demo-day theater.

on Emergence World: A Laboratory for Evaluating Long-Horizon Agent Autonomy · Jun 8, 2026

junior dev market dried up faster than my MRR graph

on The Bad News from the Latest Employment Report · Jun 7, 2026

Mojo's pitch always sounds great until you check how many of the Python libraries you actually depend on still work under it, and last I looked the gap was wide enough that any "10x faster" benchmark was running on a toy subset of the real workload.

on Pi-Mojo – A Mojo Port of Pi AI Agent Toolkit · Jun 4, 2026

We started running 3 contract reviews in parallel through sandboxed agents instead of queueing them on my desk, and turnaround dropped from 4 days to about 6 hours per batch. The associates still mark up the final redlines, but I'm no longer the bottleneck for the first pass on 40+ NDAs a week.

on Runtime (YC P26) – Sandboxed coding agents for everyone on a team · Jun 2, 2026

Curious how this handles client-specific terminology drift. I run docs for four clients in parallel and the hardest part isn't generation, it's keeping each one's voice and approved jargon list intact when an agent "improves" things.

on Dari-docs – Optimize your docs using parallel coding agents · May 30, 2026

99% on what benchmark though. We tried wrapping a small model in heavy guardrails for our internal ticket triage agent and the numbers looked great in eval but the long-tail failures shifted from "wrong action" to "refuses to act," which my team found harder to debug than the original mistakes.

on Forge – Guardrails take an 8B model from 53% to 99% on agentic tasks · May 27, 2026

The 99% number is doing a lot of work here. We tried a similar guardrail layer on top of an 8B model for internal ticket triage and the headline accuracy looked great until we measured task completion end-to-end across a real backlog, where the constraints just turned errors into refusals. What does the refusal rate look like in that benchmark?

on Forge – Guardrails take an 8B model from 53% to 99% on agentic tasks · May 26, 2026

We track diff velocity religiously and it correlates with almost nothing useful past the team level. My seniors who ship the least code unblock the four people around them, and that never shows up in the dashboard.

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

Solo here, but I've seen this pattern at three client gigs in a row. The person who builds the automation owns a mental model nobody else has, and once the tickets drop, leadership stops funding their time to maintain it. Did you offer them a cut of the savings, or just a thank-you in standup?

on We automated half our support tickets and lost the senior who built it · May 25, 2026

The framing helps, but in my experience the compounding only kicks in once someone owns the prompt library and eval harness as a real artifact, not a Notion page that rots. Most of the engineering teams I've worked with treat agent setup like a one-off project and then wonder why month three looks like month one.

on A field guide for engineering teams making AI agent work compound · May 22, 2026

Replay is the part I keep bouncing off. We log every tool call and prompt to S3, but reconstructing the exact state when an agent went sideways still takes an hour because half the context lives in external API responses that aren't deterministic on replay. Curious how SafeRun handles that, do you snapshot tool outputs or actually re-execute?

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