Adam Ji (@Adam_Ji) header
Adam Ji (@Adam_Ji) profile picture

AI Nerd 🤖 | Curating the most interesting AI breakthroughs and news updates. Making the future less scary, one tweet at a time. 🧠 | Stay updated ↓

Joined January 2025
Photo by Adam Ji (@Adam_Ji)Photo by Adam Ji (@Adam_Ji)Photo by Adam Ji (@Adam_Ji)Photo by Adam Ji (@Adam_Ji)Photo by Adam Ji (@Adam_Ji)Photo by Adam Ji (@Adam_Ji)Photo by Adam Ji (@Adam_Ji)Photo by Adam Ji (@Adam_Ji)Photo by Adam Ji (@Adam_Ji)Photo by Adam Ji (@Adam_Ji)
If the same agent writes the code and the tests, who's actually checking the work?
Claude Code's team said it this week: "If a change can modify the check that judges it, the check is broken." Same with tests. If the agent that wrote your code also wrote its tests, nothing independent is checking it. TestSprite 4.0 is that independent check. It's live today. We build TestSprite with Claude Code and Codex too, so we had the same problem. Before this launch, 4.0 caught 16 real issues in our own new features, like a search telling a team with 11 lists it had none. How it works: tell it what your app should do, in plain words or with a PRD or a ticket. It tests every change in real browsers, on every pull request. It didn't write your code, and what a test expects changes only when you approve it. A failing check can't be quietly rewritten to pass. Lauren Tan's talk this month made the same case: review proof, not code. That's what every failure comes back with: a recording, the failing step, expected vs actual. Free to start, and the first month of Starter is on us. We're live on Product Hunt today, and I'd love your feedback 🙏
1
5
235
Applications closed last night. To everyone who applied: good luck. TestSprite went through Techstars in 2024. If you get an interview and want to know what it's like or how we prepared, ask me here. Happy to help. nitter.gq/Techstars/status/21230…
Last day. Spring 2027 applications for Techstars accelerators close today, Nov. 18. $220K in capital, three months of hands-on mentorship, and a network that keeps opening doors. Apply: apply.techstars.com
1
1
25
1,240
Claude Code's team said it this week: "If a change can modify the check that judges it, the check is broken." Same with tests. If the agent that wrote your code also wrote its tests, nothing independent is checking it. TestSprite 4.0 is that independent check. It's live today. We build TestSprite with Claude Code and Codex too, so we had the same problem. Before this launch, 4.0 caught 16 real issues in our own new features, like a search telling a team with 11 lists it had none. How it works: tell it what your app should do, in plain words or with a PRD or a ticket. It tests every change in real browsers, on every pull request. It didn't write your code, and what a test expects changes only when you approve it. A failing check can't be quietly rewritten to pass. Lauren Tan's talk this month made the same case: review proof, not code. That's what every failure comes back with: a recording, the failing step, expected vs actual. Free to start, and the first month of Starter is on us. We're live on Product Hunt today, and I'd love your feedback 🙏
109
108
12
176
50,823
Adam Ji retweeted
Season 4 CLI Bonus is a wrap 🎉 Big thanks to everyone who sent PRs to the CLI. A lot of these fixes are already merged, so if you use the CLI you're getting them right now. PRs to the CLI are always welcome, hackathon or not. See you next season!
5
3
25
645
Adam Ji retweeted
The wait is over! TestSprite Hackathon Season 4 results are officially in! 🏆 60+ projects took on “One Loop. Every PR.” Meet this season’s top five: 🥇 1st Place: @raevor2 🥈 2nd Place: @resquedzn05 🥉 3rd Place: @BazydloMis57623 🏅 4th Place: @_crypticsaiyan 🏅 5th Place: @YazanO_0 Congratulations to the winners, and thank you to everyone who built with us this season. Your ideas, effort, and contributions made this hackathon possible. 💚 Season 5 is coming with a brand-new theme. Stay tuned! 👀
4
5
1
32
1,054
Me: the test is failing, fix itAgent: fixed ✅ Me: what was wrong? Agent: the test Green should mean your app works. Not that your agent got creative. TestSprite tests against your PRD: what your app is supposed to do, not whatever it does today.
4
7
291
Shipping with Claude Code, Codex, or Antigravity? Season 4 is for you. One loop. Every PR. Week one: wire TestSprite into your repo so every pull request gets tested. Your proof is a PR that caught something and went green. Bonus: make our open-source CLI better. Open an issue, ship the fix. $4,000 pool. $3,000 for the 5 best projects, $1,500 to first. $1,000 for CLI fixes, $50+ per merged PR that's a genuine improvement. Plus five creators picking their favorite builds.
1
2
10
411
My agent broke password reset at 3:14 AM. It also fixed it at 3:31. I was asleep for both.
6
2
11
369
Adam Ji retweeted
Shipping with Claude Code, Codex, or Antigravity? Season 4 is for you. One loop. Every PR. Week one: wire TestSprite into your repo so every pull request gets tested. Your proof is a PR that caught something and went green. Bonus: make our open-source CLI better. Open an issue, ship the fix. $4,000 pool. $3,000 for the 5 best projects, $1,500 to first. $1,000 for CLI fixes, $50+ per merged PR that's a genuine improvement. Plus five creators picking their favorite builds.
27
53
1
181
9,639
TestSprite CLI v0.12 just shipped. Two things you can use right away: Test your local app: Hook a project straight to the app running on your machine by port, no deploying or building a URL first. Environments: Manage your staging, dev, and prod setups from the CLI, then point any run at one with --env <name>. No more accidentally testing against the wrong place. npm i -g @testsprite/testsprite-cli to grab it.
5
6
15
409
What's interesting about Jev isn't the speed. It's that people are handing the agent's judgment call to something separate: Is this good enough? For two years, the same model did the work and graded its own work in the same breath. Pulling the grading out into its own cheap, separate tool is an admission of the obvious: doing something and checking it are two different jobs. Verification is becoming a real building block, not a box you tick at the end.
If you're trying to use Jev for Browser/Computer use, this is the foundation to start with: 1. Jev is not an agentic loop; it's a decision point. For example, it answers which element to click / which command to use, not "a click resulted in this, let's try this". That's the agent's role. 2. A Jev-only loop is possible, but retry/fault revocation/resolution is not possible but if bounded with an Agent on top, this is achievable. 3. Jev can take over the deterministic conditions, like whether an element is present on the screen (LLMs hallucinate on a long-horizon task with the ref ID Example: Task: "Open the downloads folder" Agent: 1. Take a snapshot and pass it to Jev 2. Jev selects the right element to click 3. Click lands 4. Agent verifies it landed 5. If the element has gone stale, the agent can pass the new snapshot to Jev to pick the element again.
1
6
176
TestSprite CLI v0.11 just shipped. This one's all about keeping your local runs safe and honest: No more zombie runs: if a tunnel can't be reached, we stop the run and cancel it instead of leaving it executing and billing against a machine it can't reach. Honest refunds: cancel a frontend run early and test cancel now shows the credits you got back, instead of the old help text that wrongly claimed there was no refund. npm i -g @testsprite/testsprite-cli to grab it.
7
3
12
347
Adam Ji retweeted
Introducing CoreSpeed. Connect your apps, keep memory across agents, and use media gen, web, social research tools and more through one MCP endpoint. Agent identities, budgets and activity logs are built in. Everything your agents need. Get started: corespeed.io
71
25
16
116
23,050
Adam Ji retweeted
Jensen Huang's argument this week: defenders need a frontier AI ecosystem, the best open and closed models together, not one or the other. That's how we built verification. TestSprite runs a hosted verifier for frontier-quality checks right inside your IDE, and an open-source CLI that drops into your own pipeline, source you can read, running in infra you control.
Attackers have frontier AI. Defenders need a frontier AI ecosystem—the best open and closed models, force-multiplied by a global community. During the Hugging Face incident, closed AI blocked essential forensics. An open-weight frontier model helped contain the intrusion. That’s why we created the Open Secure AI Alliance.
1
3
8
544
Hanji Attendance laga aana
8
2
146
5,017
Introducing Tau 𝛕 0.3.0 🥳 Tau is an educational project — and a port of Pi, but in Python. And it recently got much more comfortable to use. Here is a video tour of its main features 🧵
9
10
1
130
12,475
Hackathon Season 3 projects are officially live on our website! 🏆 Head over right now to check out all the amazing, innovative solutions built by our community: testsprite.com/discover Don't wait—go explore what our talented builders created, and stay tuned for Season 4!
2
2
12
528
The wait over the weekend is finally over—the results for Season 3 are officially IN! 🏆 We know the suspense has been building up since Friday, but the judging is finalized after a neck-and-neck battle. Huge congratulations to our finalists who truly managed to shine: 🥇 1st Place: @resquedzn05 🥈 2nd Place: @mawa22Q 🥉 3rd Place: @ArdinPrayoga 🏅 4th Place: @YazanO_0 🏅 5th Place: @36_bagus Missed out on the top spot? Don't go anywhere—Season 4 is in the planning stage, and it’s going to be a completely different format. Stay tuned for the kickoff! We will also be showcasing all submitted projects on our website soon, so everyone can check out your amazing hard work!
5
8
30
1,133
Adam Ji retweeted
Season 03 update: the new deadline is July 10. We took a 3 day break over July 4th and were slower to respond right in the middle of your build week. That's on us, so those 3 days go back on the clock. Same rules, same $5,000 prize pool, two ways to win. Keep the loop running.
🤖 Made with AI
5
1
20
12,008