Dexter Wang

58 posts

Dexter Wang banner
Dexter Wang

Dexter Wang

@dexterxwang

Building AI coding debugger that fixes faster → 🌟https://t.co/Iz87alXteC Thinking deeply, building boldly, sharing insights. Devs & AI folks welcome 🚀

Katılım Nisan 2016
93 Takip Edilen12 Takipçiler
Dexter Wang
Dexter Wang@dexterxwang·
@OpenAI I hope Codex can support output in HTML and diagram formats. HTML is far more expressive than Markdown. Much of Codex's text output currently feels overly abstract. HTML makes content easier to review and understand, boosting collaboration efficiency between humans and AI.
English
2
0
3
2.3K
OpenAI
OpenAI@OpenAI·
Highlights from today’s Codex Thursday launches: 1️⃣ Codex can now securely use apps on your Mac from your phone, even when your Mac is locked and the screen is off. #locked-use" target="_blank" rel="nofollow noopener">developers.openai.com/codex/app/comp…
OpenAI tweet media
English
331
392
3.9K
901.7K
Dexter Wang
Dexter Wang@dexterxwang·
We’ve observed the same. Providing LLMs with runtime context, we pushed the accuracy to 83.4%. But we found that even with the root cause identified, insufficient problem descriptions often block success.
OpenAI Developers@OpenAIDevs

The standard for frontier coding evals is changing with model maturity. We now recommend reporting SWE-bench Pro and are sharing more detail on why we’re no longer reporting SWE-bench Verified as we work with the industry to establish stronger coding eval standards. SWE-bench Verified was a strong benchmark, but we’ve found evidence it is now saturated due to test-design issues and contamination from public repositories. openai.com/index/why-we-n…

English
0
0
0
10
Dexter Wang
Dexter Wang@dexterxwang·
The secret wasn't prompting. It was Runtime Facts. We feeds dynamic runtime traces—real args, return values, and stack depth—to the LLM. It fixed 30 complex bugs that static analysis completely missed. We've open-sourced this logic as an Agent Skill: github.com/Syncause/debug…
Syncause@syncause

We achieved an 83.4% pass rate on SWE-bench Verified using Gemini 3 Pro, surpassing the 77.4% baseline. By integrating Runtime Facts, our agent observes actual execution data to fix deep logical bugs instead of guessing. Don't guess. Observe.

English
0
0
0
47
Dexter Wang
Dexter Wang@dexterxwang·
@genwinRahul Alright, it seems I'm the fish that took the bait. You must know about this matter😭
English
1
0
1
7
Dexter Wang
Dexter Wang@dexterxwang·
@genwinRahul lol, others can't open this web because it is "localhost". This means it's a local address accessible only on the machine where the web service is running, rather than being publicly available
English
2
0
1
15
Rahul
Rahul@rahulgrows·
OpenClaw is fucking insane i know literally NOTHING about coding. ZERO. and i just built a fully functioning web app in minutes. http://localhost:5173/ check it out
English
1
0
1
157
Dexter Wang
Dexter Wang@dexterxwang·
Everyone is discussing 'Agent Skills' right now. It is indeed an efficient paradigm for modular reuse. We released the Debug Skill to apply this model to a specific problem: reliability. By standardizing the debugging context, we aim to improve fix accuracy and enable the AI to deliver usable code in the first iteration.
Syncause@syncause

Introducing Syncause Debug Skill and MCP Server. This integration allows AI Agents to automatically retrieve runtime execution data before proposing code fixes. It enforces a "Mandatory Forensics" workflow where the AI must cite specific runtime evidence. github.com/Syncause/debug…

English
1
0
0
24
Dexter Wang
Dexter Wang@dexterxwang·
Why can’t the ChatGPT or Gemini web UI render Mermaid diagrams directly? Grok/DeepSeek can do it just fine. Getting raw Mermaid code is a pretty bad UX. Rendering Mermaid isn’t technically hard, and it’s MIT-licensed anyway. So what’s the real reason? Security? Product choice? Anyone know?
Dexter Wang tweet mediaDexter Wang tweet media
English
0
0
7
183
Dexter Wang
Dexter Wang@dexterxwang·
I’ve seen the same pattern. People often assume that once AI gets “good enough” at coding, debugging will just disappear. But bugs often come from shifting requirements breaking hidden assumptions — and AI isn’t immune to that. That’s why I’m building a tool to make AI better at debugging.
Santiago@svpino

68% of developers say they spend more time debugging AI code than writing new code. (According to a survey of 500 software engineering leaders and practitioners.) AI coding assistants help us write code faster, but they increase the amount of insecure and broken code we ship.

English
0
0
0
8
Dexter Wang
Dexter Wang@dexterxwang·
I spent an hour debugging a complex, real-world project with Codex, just mentally parsing code and burning through tokens. It was a total nightmare. Then, I captured a runtime stack trace and fed it directly into the context. The investigation path became instantly clear, and we pinpointed the root cause in no time.
English
0
0
0
10
Dexter Wang
Dexter Wang@dexterxwang·
On Windows, after configuring MCP for Antigravity, I ran into the "Error: calling "initialize": EOF". It's due to a missing PATH environment variable. Fix: Add the PATH to the env parameter in your MCP config. Steps below 👇 Run this in PowerShell to get the paths: Get-Command node | Select-Object -ExpandProperty Source It'll output something like: C:\Program Files\nodejs\node.exe Then add it to your mcp_config.json:
Dexter Wang tweet mediaDexter Wang tweet media
English
0
0
0
45
Dexter Wang
Dexter Wang@dexterxwang·
@saen_dev @syncause Exactly, I have to learn GTM from scratch. It's tough from a developer's perspective because I need to learn how to communicate in a whole different way with users.
English
0
0
0
9
Saeed Anwar
Saeed Anwar@saen_dev·
@dexterxwang @syncause This hits hard. Most technical founders (myself included) default to "I'll just build a feature for that" when the real problem is distribution. Code is the easy part. Getting people to care and pay is the actual startup game.
English
1
0
0
16
Dexter Wang
Dexter Wang@dexterxwang·
1/ Over the past decade, my identity has shifted 4 times: Junior Dev ➡️ Tech Lead ➡️ Product Manager ➡️ Founder. I used to think this was a linear upgrade path. I was wrong. Becoming a Founder wasn't an upgrade; it was a complete teardown and reconstruction of my reality.
Dexter Wang tweet media
English
1
0
0
32
Blake Emal
Blake Emal@heyblake·
Drop your project URL Let’s drive some traffic
English
1K
24
554
63K
Dexter Wang
Dexter Wang@dexterxwang·
6/ This path isn't a straight line upward. It’s a cycle of rebirth. Every perspective shift requires killing a part of your past self. If you are transitioning from Engineering to Business, don't look back. The real world is messy, but it's worth it. #Startup #SaaS #BuildInPublic
English
0
0
1
16
Dexter Wang
Dexter Wang@dexterxwang·
5/ The most painful realization? As a technical founder, my instinct is to "write a line of code" to solve every problem. But 90% of startup killers—distribution, pricing, psychology—cannot be fixed with a git commit. Building @Syncause, I fight the urge to open my IDE every day.
English
2
0
0
28

Keşfet

@OpenAI @antigravity @saen_dev @syncause @heyblake @elonmusk @BarackObama @taylorswift13