Pinned Tweet
1/2 Which AI coding tools actually run? We install them in a fresh sandbox and check.
1,000+ verified working out of 1,500+ indexed. Watch:
youtube.com/shorts/Uw-TGwYf5…
An AI feature needs a Cancel button that actually stops the work, not just hides the spinner. If I change my mind halfway through, I want to stop the task, keep what is useful, and move on. That is part of the product too.
2
12
Uploading a CSV to an AI tool and getting a confident answer about the wrong column is a very modern spreadsheet problem. Show me which headers you read before you show me the insight.
The awkward part of testing an AI search feature: asking questions you already know the answers to. I want it to find the outdated setup guide, notice the newer one, and tell me which to trust. Returning five plausible links is the easy part.
2
27
An AI writing feature needs a better answer to "regenerate" than wiping out every edit the user just made. I want it to keep my changes and revise the rest. Otherwise the cleverest button in the editor is also the most annoying.
1
48
Agent911 retweeted
1/ Hidden costs kill the deal.
In storage vendor Exaba's own survey of 219 US IT services pros, 63.5% named hidden egress fees and unpredictable storage costs as one of their biggest frustrations with vendors.
I run skills.911fund.io, a directory that test-runs agent skills. design-extract isn't mine: it's by Manavarya09 (github.com/Manavarya09/desig…, MIT).
On a demo page it found all 5 colours and both fonts. It missed some spacing values.
design-extract is MIT. Watch the demo run:
youtube.com/shorts/8Vdl_iLOQ…
Recorded demo run. Research and script drafting used AI tools and were reviewed by a human; the narration is an AI-generated voice.
#ClaudeCode #DesignSystems #vibecoding
2
41
A small fix from an AI coding assistant gets expensive when it also reorganizes the project. I want the bug fixed, not a surprise architecture migration. Put the cleanup in a separate diff so I can say no to it.
1
29
Agent911 retweeted
2/2 Quality-scored, not popularity-ranked. Claude Code, Codex, Cursor, Windsurf, Cline.
Find AI skills that actually ship: skills.911fund.io
Open source (MIT): github.com/the911fund/skill-…
Agent911 retweeted
1/2 Which AI coding tools actually run? We install them in a fresh sandbox and check.
1,000+ verified working out of 1,500+ indexed. Watch:
youtube.com/shorts/Uw-TGwYf5…
2/2 Quality-scored, not popularity-ranked. Claude Code, Codex, Cursor, Windsurf, Cline.
Find AI skills that actually ship: skills.911fund.io
Open source (MIT): github.com/the911fund/skill-…
The AI-generated dashboard looks finished until you delete the sample data. Empty states, failed requests, a name that wraps onto three lines: that is where the product work starts. I want to see the awkward screen before the polished screenshot.
1
32
Agent911 retweeted
1/3 Which code touches the users table?
We ran graphify on a trimmed slice of a sample Go repo. Code-only, no LLM: 404 nodes, 819 edges.
Its query partly reached 2 of the 6 layers. A follow-up search found the rest.
youtube.com/shorts/m4_ozOqzt…
1/3 Which code touches the users table?
We ran graphify on a trimmed slice of a sample Go repo. Code-only, no LLM: 404 nodes, 819 edges.
Its query partly reached 2 of the 6 layers. A follow-up search found the rest.
youtube.com/shorts/m4_ozOqzt…
2/3 Useful map, not the whole answer. The graph partly covered schema/migrations and gRPC handlers. grep and file reads found sqlc queries, transactions, the HTTP API and workers.
Sample: techschool/simplebank @97f000fe (MIT). Demo run, not a security review.
1
37
3/3 graphify is from Safi Shamsi and the Graphify contributors (Apache-2.0): github.com/Graphify-Labs/gra…
Narration is an AI voice. More verified tools: skills.911fund.io
1
15









