Steven Fang

This will be a great session with Josh Smith and our AWS partners on turning AI into real enterprise value—not just isolated pilots. It’s grounded in lessons we’ve earned the hard way, figuring out how to build internal capability that compounds and creates measurable impact across merchandising, supply chain, and customer experience. At the center of it all: prioritizing, governing, and operationalizing AI in a way that actually sticks. The session walks through the frameworks and decisions that have made the difference. If you’re at re:Invent and want a clear view of what it truly takes to scale AI across the business, this is the one to catch.

8

Satya Nadella (Microsoft CEO) made a point this week on a topic I’ve discussed and thought a lot about recently: https://lnkd.in/eyXQvTdt I like his human-capital / token-capital framing though I’d push it one step further. Most of the conversation still assumes frontier capability stays scarce. It won’t. What required access to a state-of-the-art model 18 months ago increasingly runs on open weights and hardware you own. For a large share of the work that creates real enterprise value, the absolute frontier is no longer the constraint. And once everyone has the capability, the capability stops being the advantage. What’s left is harder to copy: knowing what matters, measuring it, and improving it in a loop. The organizations that win won’t be the ones with the best models. They’ll be the ones that learn faster than their competitors. That requires closing the loop between decision, outcome, and feedback. AI brings scale and speed. People bring judgment, perspective, and accountability. Neither closes the loop alone. The hard part was never deploying AI. It’s building the enterprise measurement system that lets you learn from it.

8

Garry Tan says MCP sucks. Perplexity is moving to direct APIs. Everyone's surprised. I'm not. We tried something similar with our RAG integration layer. Clever abstraction, clean architecture on paper. 3 weeks later: context bloat, brittle auth, more debugging than building. Replaced it with a direct API call. Took 2 hours. Has run cleanly for 4 months. In production, boring infrastructure scales and clever infrastructure debugs. So, build the simplest thing that ships first.

30

4 Comments

Week 3 after Amazon and the momentum is starting to build. This past week was full of forward motion. I created and launched an investor pitch page, attended SoCal Startup Day, made some great connections, and was unexpectedly invited to pitch my startup idea at a competition next month. I’ve also had a few promising phone screens for AI-first engineering leadership roles. It feels like doors are opening in several directions at once. In parallel, I’ve been heads-down making progress on my MVP. The pain I’m exploring is one that many engineers are starting to feel: AI can generate code incredibly fast, but it doesn’t yet generate reliable systems. It creates components that look complete, and sometimes even functional, but they often lack the deeper design considerations that make software sustainable even as it evolves. Metrics and observability, long-term evolution, lifecycle, cohesion across layers. AI solves for the prompt, not the vision and not the architecture. And often vibe coding creates overconfidence that a real system has been created, when what’s actually been produced is only a re-invinted wheel prototype with missing spokes. That gap fascinates me. It feels like we’re still applying pre-AI assumptions to an AI-first world. Principles and patterns shaped around human limitations shouldn't be automatically mapped to an environment where code can be produced, rewritten, or replaced instantly. The more I experiment, the more curious I become about what becomes possible when we stop assuming the baked-in human constraints that shaped the last 20 years of software engineering should define the next 20. This week I validated a key part of my thesis in code. I won’t go into details yet, but it gave me confidence that the direction I’m pursuing isn’t just theoretical. There is a real path forward here. If you’re exploring how AI is reshaping the foundations of software development, or if you’ll be at any upcoming startup or pitch events in Southern California, I’d love to connect! #startups #founders #ai #sdlc #architecture #preseed #socalstartupday

32

GitLab CEO Bill Staples, AI engineer Angelo Rivera, and Senior Director Sarah Waldner unveil GitLab 18.4, revolutionizing DevSecOps with AI-powered innovations. Featuring lightning-fast Knowledge Graph, intelligent Agentic Chat, automated pipeline fixes, AI Catalog, and enterprise-grade governance.

6