Stephen Han

usually passes the Turing test // have courage and be kind // please excuse my pre-coffee tweets

[Stephen Han @stephen__han]

WHAT IF: The future of software is significantly more secure. The idea of shift-left can reach new heights because we have newfound bandwidth to automate the bottom 80% of issues.

[Stephen Han @stephen__han]

Still need human review but the bottom 80% of quality/security cases can basically already be automatable for the cost of tokens.

[Stephen Han @stephen__han]

The future of LLM audit and observability is probably at least 100x what it is now.

[Stephen Han @stephen__han]

3 Key Areas to Watch:
1. Traces (think Langfuse, Arize)
2. External actions and integrations
3. Artifact and code generation

[Stephen Han @stephen__han]

It's how we survive and navigate the world.

[Stephen Han @stephen__han]

Humans are probabilistic.

[Stephen Han @stephen__han]

Literally we are pattern matching beings.

[Stephen Han @stephen__han]

There's basically no world where Jevon's paradox ceases to be true.

Same for Moore's Law. Maybe the base number changes and the shape of the industry shifts. Either way the exponent carries the weight.

[Stephen Han @stephen__han]

Conviction >

[Stephen Han @stephen__han]

@petecodes Interested in the next one as well

[Stephen Han @stephen__han]

And my advice is always answer the question; Better that than to ask it all your life.

[Stephen Han @stephen__han]

Plus fascinating takes on defense tech applications and why the future of advertising isn't demographic bucketing - it's individual creative generation in real-time.

[Stephen Han @stephen__han]

🎙️ New podcast with @chetmancini about the practical realities of implementing AI in production systems.

https://t.co/0tvyIafWgL

[Stephen Han @stephen__han]

Key insights:
→ Clean data beats fancy prompts every time
→ AI compliance is the next big vertical
→ Personalized ads will be generated per individual
→ Engineering fundamentals still matter at higher abstractions

[Stephen Han @stephen__han]

Key insights:
→ Small AI models beat large ones for specific use cases
→ Social science students can build startups easier than ever; and they bring a unique edge
→ Is traditional VC dying in the AI era?
→ Build for your community, not the world

[Stephen Han @stephen__han]

🎙️ New podcast drop: From FBI counterintelligence expert to serial entrepreneur

@MikeSJacobs1 is a serial entrepreneur and founder of Berkeley's startup incubator for cognitive science students.

https://t.co/4CM3FWFmOF #AI #Entrepreneurship #Startups #Patreon #TechPodcast

[Stephen Han @stephen__han]

In this episode, Mike shares his unique journey from FBI counterintelligence to ideating on Patreon and cofounding OrderMark, and discusses why the future of AI entrepreneurship lies in small, focused applications.

[Wyatt Benno @wyatt_benno]

1/ We made verifiable AI (zkML) 3-7x faster than everyone else! Here's how.

We asked a simple question. What if we modified a16z Crypto's sumcheck based & lookup-centric zkVM "JOLT" for ML operations?👇🧵

[Stephen Han @stephen__han]

Also, check out Fred's photography(!): https://t.co/UfpYMkjuUu

#ProductionAI #AIAgents #DataScience #MachineLearning #TechPodcast #EnterpriseAI #AIEvaluation #SoftwareDevelopment

[Stephen Han @stephen__han]

I got to chat with Fred Yang about his experience designing and building enterprise grade production agents at @indeed .

Interview: https://t.co/tzJcRZDNAt

[Stephen Han @stephen__han]

We talked about:
-Why multi-agent architectures might be over-engineered for most use cases
-How evaluation-driven development is becoming the new standard for AI
-How the cost dynamics of code writing vs. code reviewing are fundamentally changing