Two Viral Posts Prove the AI Bottleneck Isn't the Technology — It's You.

If you've hit a wall with AI — using it consistently but still getting output that needs too much repair — the problem almost certainly isn't the model. Two posts landed on the front page of Hacker News in the same week in May 2026, with a combined 1,374 points, and neither was about a model release. One was about what happens before you prompt. The other was about what happens after. Both pointed at the same person.

What two viral posts said in the same week

On May 17, Frederick Vanbrabant published "AI Won't Speed Up Your Processes" — 678 points. His argument: AI cannot fix a slow process, because the bottleneck is never the model. It's upstream — vague requirements, incomplete documentation, unclear success criteria, problems that haven't been defined clearly enough to hand to a person, let alone a language model. He noted that the "built it in three days with AI" stories routinely omit the 40-plus days of specification work that actually gate the project. Feed it a muddy input and you get a muddy output at higher velocity.

Four days later, "No Slop Grenade" reached 696 points, naming the failure at the other end: pasting walls of AI-generated text straight into Slack messages, client emails, and shared documents without editing. The complaint was blunt — recipients now have to compress and verify the text themselves, doing the work the sender skipped.

Simon Willison flagged a third version the same week: Armin Ronacher's critique of AI-generated reports that sound authoritative but are speculative. Confident tone, guessed diagnosis. The output reads like it knows. Often it doesn't.

Three failure modes, one underlying pattern — treating the model as both the input processor and the output quality gate. It's neither.

The two seams that matter

The AI skill that compounds isn't prompting technique. It's the work at two seams.

Before: scope it. Write a brief — three lines is enough. Who the output is for, what decision or action it needs to support, and what format it should take. Vanbrabant's point isn't that AI is useless for hard problems; it's that a vague prompt amplifies whatever ambiguity is already in the process. Tightening the input reduces it at source.

After: edit it. Would you send this exact text if you'd typed every word yourself? If not — if you'd cut the third paragraph, tighten the opening, or delete the sentence beginning "it is worth noting that" — it isn't ready. This isn't distrust of AI. It's the gap between what a model produces and what actually serves the person receiving it.

Dara: the upstream problem

Dara is content and campaigns lead at a 45-person SaaS company, using AI to draft competitive comparison pages. Before she started scoping explicitly, the output was generically accurate but useless — correct facts, no specific differentiation, no language tuned to a technical buyer evaluating procurement risk.

Her fix was three lines written before every brief: the buyer role, the specific objection the page has to handle, and the competitor claim she's countering.

The output stopped being generic the moment the input did. She didn't change models. She changed what she handed to the model.

Kwame: the downstream problem

Kwame is an operations lead at a 90-person consulting firm, sending weekly status updates to three clients. For six weeks he pasted AI-generated summaries straight into client emails.

Then a client flagged one. The summary described a project risk as "being actively monitored" — a phrase the model produced, not a fact Kwame had verified or even intended. The update sounded authoritative. It wasn't.

He now does two passes before anything leaves his outbox: one to verify factual claims, one to cut anything he couldn't personally defend if asked. It adds five minutes per update, and it removes the version of him that sends things he hasn't really read.

The one thing

A better model given a vague brief produces a better-sounding version of the wrong thing.

The upstream work — defining what you need clearly enough to ask for it — and the downstream work — editing before it reaches another person — are still yours. Both viral posts said the same thing from opposite ends: the model wasn't the problem. The work at the seams was.