The strongest argument Convex has made this year, that nine out of ten agent-built Postgres apps silently corrupted data and none on Convex did, is served from a subdomain named after a different product. The Product Marketer is hired because Convex ships but does not tell people.
First action
Check the repo and load tests can be run by an outsider in under an hour.
How to measure it
Signups attributed to the research page
- Now
- –
- Target
- 1,000
- By
- 60 days
Steps
Check reproducibility
An engineer outside the project runs the tests from the README.
Software Engineer, AI 1 day
Move the page
A stable convex.dev URL with a redirect from the subdomain.
Creative Director / Head of Design 1 day
Write the Stack post
Method, results, limits and what it does not prove.
Product Marketer 2 days
Link it in
Homepage, compare pages and transaction docs.
Head of Developer Experience 2 hours
Launch
Founder thread, Hacker News by the author, Discord announcement.
Product Marketer 1 day
Answer critics
Engineers reply to every serious challenge in public.
Software Engineer, AI 1 week
Measure
Signups, links and discussion threads.
Product Marketer 1 hour
Why it works
Move the research to convex.dev/research/agents-break-postgres, link it from the homepage, the Supabase and SQL compare pages and the docs on transactions. Launch it properly: a Stack post with the method and repo, a short founder thread, a Hacker News submission written by the engineer who ran it, and the raw data downloadable.
It is substance, not hype: reproducible numbers about a failure developers fear, presented the way skeptical engineers trust. That is exactly the bar the Head of Marketing brief sets.
Risks and how to handle them
Postgres experts call it unfair
Publish the prompts and invite them to rerun with their own guidance.
Evidence
- Your websiteResearch page "The backend platform your agents can't break" (served at chef.convex.dev): frontier coding agents one-shot the same ten apps on Postgres and on Convex; "9 out of 10 apps silently corrupted data on Postgres. None of them did on Convex." A shopping-cart app on Postgres falls to 28% correct updates at 256 concurrent shoppers while Convex holds at 100%.Source
- HiringProduct Marketer: "We ship frequently, but aren't great at talking about it"; "Once developers find us they tend to stick around"; "we'd sooner publish a 'how to give Convex less money' post than pretend what we do is magic"; owns launches end to end and writes the post and cuts the assets; "connective tissue between engineering, developer experience, design, and creators".Source
- HiringHead of Marketing: "At a company our stage the founders (James and Jamie) are the de facto CMO(s)"; own positioning, launches, content, community, events and brand; "Decide what we don't do. Pick the two or three motions that matter most and go deep"; "nothing hypey, nothing generic, nothing we'd be embarrassed to put in front of a skeptical engineer"; "Ship a bunch of cool marketing nobody would believe came from a database company."Source