staff engineer @ nourish // ex-loom // hot takes are my own
Joined June 2019
- Tweets281
- Following578
- Followers59
- Likes46.5K
ivan martinez retweeted
This Jensen Huang thing is so ugly. Guy takes a company from zero to absolute pinnacle, through blood, sweat, and tears, a pure founder’s founder, great products, makes so many people rich beyond their imaginations, champions open free local AI, and then the incumbent modern managerial safetyist human instrumentality establishment collectively decides they want to rip it all away from him because they think they can “do better”.
shouting this from the rooftops
The rise of the AI Platform Team.
After speaking with hundreds of CIOs, CISOs, and Heads of AI, I’m seeing a new operating model emerge: centralize the platform and governance, embed internal FDEs in the business.
Uber was the first to do this and and the whole market is following suit.
1) AI Platform Teams are becoming the gatekeepers and enablers of enterprise AI.
Staffed with Engineering, IT, and Security talent, these teams evaluate vendors, integrations, agent architectures, and deployment options. They establish the approved path to production so every business unit doesn’t have to repeat the same technical and security reviews. The best teams will make it faster to ship safely.
2) Enterprises are hiring FDEs inside every business unit
Business units need full-time FDEs to deliver Tier 1 use cases. These FDEs are tasked with understanding business workflows, identifying opportunities for automation, and owning delivery. I expect demand for these roles to grow sharply, not just at AI vendors and consulting firms, but now inside every enterprise. The platform team creates the paved road. Embedded FDEs turn it into business impact.
3) Production governance is the prerequisite for scale.
In 2026, many enterprises are working to get their first agentic automations into production. In 2027, I expect the challenge to shift toward operating them at scale. Imagine thousands of apps and agents accessing business systems around the clock. Who can access what? What happens when an agent loops, exceeds its budget, or attempts an unauthorized action? Who can stop it immediately? Access controls, rate limits, cost caps, kill switches, and defenses against prompt injection become part of operating the workload. Getting one agent into production is a milestone. Governing thousands requires systematic operation via platforms.
The leaders of AI labs love to discuss the jobs AI will kill. We should pay more attention to the roles AI is creating:
AI Platform. AI Enablement. Internal FDEs.
30
vibe coded a betting platform and 2 hours later it was in production
Vibe coding is a generational scam that sells people the fantasy of bringing their own idea to life without knowing how to code. The reality is that production software requires real software engineering knowledge and practices
1
27
i always felt like such a dolt for not being gods gift to frontend by knowing every single react api / interface / dissertation-level knowledge for frontend
but i was always curious about how systems work together
now there's sota models that do all of the little things for me and i can focus on building systems instead
i love the world we're in
35
not every implicit state machine is worth fixing btw
all that time spent pruning tiny weeds when your tomatoes need tending to
1
102
I am the software factory
the Clanker is just the assembly line
killed my software factory. don't hold onto anything.
rex + long running claude sessions that delegate to subagents with specific roles is all i need.
more malleable and less likely to pollute or overcomplicate matters. opus 5.5 does not need ceremony.
25
