AI/ML. Worked at Meta, Uber, Amazon, Apple, and Microsoft building apps, developer platforms, and hardware. Tweeting about LLM psychotherapy.
San Francisco, CA
Joined August 2008
- Tweets21.6K
- Following428
- Followers6.4K
- Likes651
I'm sorry, did claude code just offer I write 5-10 lines of code? Ha?
"Optional, if you want to write some of the code yourself: ... it’s 5–10 lines."
Wow, the new Flux 3 Max being offered at 25% for the next month sees really cost-effective.
Side-by-side comparison of Flux 3 $3.19 vs. @fal Flux 3 Max $0.79 performance on the same prompt.
Justin Angel retweeted
Excited to share a talk from @aiDotEngineer SF World's Fair on Building Blocks for Uber’s Software Factory that @udaykiran and I did back in June 2026.
At @UberEng we’ve focused on creating modular infrastructure for the agentic era, including Context Graph, Model and MCP Gateways, Cloud Developer Environments (CDEs), and an AI Assistant.
These foundational building blocks enable us to build an Agentic SDLC (aka Software Factory):
Spec-Driven Ideation: AI Assistant (Cortana) pulls past experiments, usage data, and team history from our Context Graph and turns a Slack thread between Product, Design, and Engineering into a structured spec with design variants.
Autonomous Code Generation: Minion, Uber’s Cloud Coding Agent, plans across repos and delivers PRs directly from that spec, in an interactive mode or a fully autonomous mode.
Self-Improving Validation: Every PR runs through a continuous validation loop: automated tests, visual validation, self-healing CI, and an AI code review. All before a human spends any attention on it.
Managed Maintenance Loops: Skills enrolled against specific services and feature paths run weekly to clean up dead code and optimize the codebase, then improve themselves based on review comments left on their own diffs.
With 70%+ of PRs now originating from agents, our engineers are shifting from manual execution to orchestrating agentic workflows across the development lifecycle.
Watch the full talk 👇
youtube.com/watch?v=17-YSUHo…
Justin Angel retweeted
Been Watching, Learning , and using the tools Justin provides Amazing content @JustinAngel
youtube.com/watch?v=vXiB0UdD…
Justin Angel retweeted
did a deep dive on how AI is affecting the mental health care space. surprised to learn it's not all doom and gloom sfgate.com/tech/article/arti…
Wow, the emotional expressiveness of Seedance 2.5 720p is wild. The facial micro-expressions, the hand gestures, the eye gaze. It's all so much more crisp.
Going to burning man? Have a Claude account? I'll use your weekly limit when you're gone.
Like a plant / dog / house sitter, but for tokens.
Amazing. @ElevenLabs provided a grant to operate PracticeTherapy.ai voice coach for free for the next year.
We can keep it free for mental health professionals to train in psychotherapeutic modalities. Thrilled.
Testing MiniMax H3 quantized (int8) vs Seedance 2.0 full for use in PracticeTherapy.ai videos.
🤔
What's a fused kernel?
The real cost of GPU compute isn't the compute itself, it's shuffling memory from CPU<=>GPU. So if we can do X operations on the GPU in one roundtrip, that's faster than the same X operations in many roundtrips.
Fused kernels take unoptimized code executing in multiple GPU kernels, and execute it on the GPU in one kernel.
GPU programming is so hawt right now. Here's an intro in the context of building your own LLM @ go.JustinAngel.ai/video-5
Tested video Auto-Dub on @ElevenLabs English-> Portuguese -> English.
Can't say I'm impressed with either of these. The lip-syncing is bad.
Opus 5 Max is a hill-climbing super power. This is SWE L5-level FAANG work easily.
I gave it a system that had ~1000 latency reports with many segments with a prompt on how to "make it faster".
P90 3.6s, P50 2.6 -> P90 1s, P50 0.9s.
It made it so fast I had to introduce a delay in the UI.
In 18 years on twitter, I’ve never once quote retweeted. This is important.
I find great reassurance in sitting with Bob’s words.
We’re not vibe coding, this is just what 95% coding looks like for 95% of projects nowadays.
Software engineers aren’t being replaced, we’re becoming more productive.
I’m significantly older than you. I started coding in the late 60s. My current strategy is to not read any of the code written by my agents. That’s the only way I can take advantage of their productivity. What I do instead is to surround the agents with extreme constraints. Unit tests, gherkin tests, QA procedures, quality metrics, mutation testing, test coverage, and a plethora of others. In the end, I have very high confidence in the code they produce because they’ve had to run the gauntlet of all of my constraints and tests.
🔥 New website: PracticeTherapy.ai
Free training for mental health professionals.
Plays client videos, listens for skill criteria, and provides voice feedback. All AI-supported features following evidence-based practices.
