Quick gut check I've been using while building AI systems:
- Are we storing everything or only what matters?
- Can the system update what it "knows" about a user over time?
- Can that knowledge be inspected and corrected?
- Do our evals catch edge cases and regressions?
The hard part is not generating responses.
It's building systems that learn in a reliable way.
The hardest part of building AI agents isn't the AI.
It's making them predictable.
Users expect:
- consistent answers
- reliable actions
- no weird edge cases
Getting there means:
- strict tool schemas
- defensive prompts
- evals for edge cases and regressions
- retries and validation
- logs for everything
AI feels magical in demos.
Production is engineering.
🚨BREAKING: Microsoft just solved the "Agent Loop" problem.
Agent Lightning is an open-source framework that lets agents learn from their own mistakes using Reinforcement Learning.
Your agent fails a task → Agent Lightning analyzes why → Updates the prompt automatically → Next run succeeds.
100% Opensource.
@widgetflow Hey AJ! We're @madebymonogram, a US-based company. We're a Vercel Certified Partner specializing in modern web apps and composable platforms. Shooting you a DM.
@AJCentrone Hey @AJCentrone! We're Monogram. Vercel Certified Solution Partner focused on composable web platforms. Build fast, scale easily, your team can update content without devs. Happy to chat if you're still looking. https://t.co/nN1Ic2EJ1O
@Shwinnabego We're a design + engineering shop in Atlanta focused on composable web platforms and AI-enabled experiences. Not social-first, but strong on the web dev/modernization side if any of those requests fit. Happy to chat if useful → https://t.co/nN1Ic2EJ1O
We now support Bun on Vercel.
Bun is an open runtime pushing the frontier on DX and performance.
Start building and see our latest benchmarks.
https://t.co/WvE5sQzxe0
Gemini Flash for IVR is wild. 🤯
~75% faster on big prompts, latency barely changes as context grows.
For real-time, natural conversation, nothing else comes close.
#gemini#ai#googleio
Legends, old friends, and new faces—Interrupt 2025 delivered. A full day of honest conversations on building AI agents, what’s working, what’s not, and how to get to prod. Shoutout to the @langchain crew for a great event.
2025 will be the year of agentic systems
The pieces are falling into place: computer use, MCP, improved tool use. It's time to start thinking about building these systems.
At Anthropic, we're seeing a few best practices emerge - we wrote a blog post with our findings:
Just wrapped up an amazing offsite in Mexico City 🇲🇽 🌮 — thriving at the intersection of Design + Engineering + Service depends on a team sharing strong friendships, laughter, good food, and a sense of wonder.
¡Hasta luego, #CDMX!
🤩 Interested to learn more about leveraging Gemini's advanced image recognition and expansive context window? See how @googledevs partners #BuildWithGemini 🛠️
Deep dive into Gemini 1.5 Pro explorations with @madebymonogram ↓ https://t.co/E02kWOo4sM
My richest mentor told me...
"I win more for one reason. I move fast. By the time most people are done analyzing, I've already made three mistakes and found a better way."