Two Viral Posts Prove the AI Bottleneck Isn't the Technology — It's You.
Two Hacker News posts hit the front page the same week with a combined 1,374 points. One named the upstream problem — bad specs going in. One named the downstream problem — raw AI output going out. Both point to the same fix.
By Forge Team
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 upvotes, and neither was about a new model release. They were about what happens before you prompt and what happens after you get the response. Both pointed at the same person as the bottleneck.
What two viral posts said in the same week
On May 17, Frederick Vanbrabant published "AI Won't Speed Up Your Processes" on his personal blog. It reached 678 points on Hacker News. His argument: AI cannot fix a slow process. The real bottleneck isn't execution speed — it's the specification work that precedes any generation. Vague requirements, incomplete documentation, unclear success criteria. He pointed out that the "AI development in 3 days" narrative ignores the 40+ days of upstream work that actually stalls most projects. Feed AI a muddy input; get a muddy output at higher velocity.
Four days later, "No Slop Grenade" hit 696 points. The post coined a term for a specific workplace behaviour: pasting large walls of AI-generated text directly into Slack messages, emails, or shared documents without editing them first. The argument was that it wastes the recipient's time, kills productive back-and-forth, and signals that the sender hasn't done the thinking that would make the output useful. Two days later, Simon Willison flagged the same pattern in technical communication — AI-generated issue reports that sound authoritative but substitute confident tone for actual observation, leaving readers unable to distinguish diagnosis from speculation. (Willison, simonwillison.net, May 24.)
Two posts, same week, same diagnosis from opposite ends of the same process.
The two seams that matter
The AI skill that compounds over time isn't prompting technique. It's the work at two seams: before and after.
Before: scope the task precisely enough that what comes back is editable on the first pass. That means specifying the format, the length, the audience, what you already know, what the recipient will do with the output. The more constraints you give, the less repair work lands on the other side.
After: edit what you get down to what you'd sign your name to. The test is simple — would you send this exact text if you'd typed every word yourself? If not, it isn't done. This isn't about distrust of AI. It's about the gap between what AI produces and what actually serves the person receiving it.
A product manager's brief that wasn't
A product manager at a 35-person fintech startup needs to brief a freelance copywriter on four landing page variants for a new savings feature. She prompts: "Write a creative brief for landing pages about our savings feature." She gets 500 words back and forwards it to the copywriter.
The copywriter replies with nine questions — all of which were answerable before any AI was involved. Target customer. Desired action. Tone constraints. Competitor points to avoid. The session runs 45 minutes instead of 10. The AI produced something that looked complete. It wasn't, because the prompt didn't specify what "complete" required.
The brief was the work. The prompt was too thin to produce a brief.
Write the constraints before you write the prompt. The brief is the work.
An HR report nobody could read
An HR business partner at a 90-person professional services firm uses AI to summarise 60 employee survey responses into a slide for the leadership team's quarterly meeting. The AI returns eight bullet points, each two sentences long, covering every theme at equal weight.
She sends it as-is. In the debrief, three of the four leaders say they couldn't identify the headline finding. The AI treated all themes as equivalent because she hadn't told it otherwise. The top concern — workload distribution — had been flagged in the previous two quarters and carried more weight than the others. That context lived in her head. She needed to put it in the prompt or cut it into the output before it left her hands.
The edit would have taken five minutes. The confusing slide took longer to undo.
Cut the AI draft down to what you'd defend in a meeting.
AI generates at speed. What you contribute is precision going in and judgment going out. Both viral posts said the same thing with different examples: the model wasn't the problem. The work at the seams was.
Fix a vague prompt before it generates output you can't use.
Like this post?
Get the next one in your inbox. Practical AI skills, no filler.