Dawe

Founder & CTO of NOMO
luminary, delusional optimist, paradigm provocateur exploiting game theory, sapient savant

[Dawe @Dejvyq]

Writing boilerplate was never the hard part. Knowing which system architecture won't collapse under load in 6 months is

[Dawe @Dejvyq]

AI didn't make junior developers obsolete, it made senior developers 10x more leveraged

[Dawe @Dejvyq]

The actual bottleneck in production AI is deterministic reliability. Nobody cares about a 2M token context when the agent hallucinates a tool argument.3. Code generation reality

[Dawe @Dejvyq]

Everyone is obsessing over context windows and reasoning tokens

[Dawe @Dejvyq]

The gap between closed frontier labs and open weights isn't closing, it's just shifting. Labs own the massive inference clusters, but open source owns the developer ecosystem and local tooling. Distribution beats raw MMLU.2. Real agent bottleneck 😅

[Dawe @Dejvyq]

Most agent startups are building better prompts
Almost nobody is building the execution spine

When an autonomous agent can call Stripe, AWS, and production DBs:
"Please behave" in the system prompt is a liability, not security
Agent identity and reversible blast radius are btckl

[Dawe @Dejvyq]

Most AI agent security is a prompt telling the model to behave

[Dawe @Dejvyq]

An AI agent without identity is just a process with API keys. Every agent needs distinct identity, least privilege, and auditable delegation. Regex isn't security architecture: trust boundaries must be enforced by IAM, not a system prompt

[Dawe @Dejvyq]

@vinj_ai My own build tool. If you want to try out and give me a feedback I could send you my repo

[Dawe @Dejvyq]

The moat isn't having AI. The moat is knowing what to never hand over to it."

[Dawe @Dejvyq]

AI won't be the biggest threat to jobs. The real threat is the company that adopts AI faster than competitors, only to realize they automated away their own judgment

[Dawe @Dejvyq]

Week 6 building the agent control plane:
- 28 production writes gated by a policy check
- 10 MCP connectors working end to end
- 6-second credential lifetime
- lot of things that broke that I did not expect
Thread on what broke below

[Mark Zuckerberg @finkd]

Introducing Muse Wonder, our new family of frontier models with state of the art performance for personal and professional agentic work.

[Dawe @Dejvyq]

True

[Money Talks @MoneyTalksTwt]

Kill the belief that getting rich has to be hard

[Elon Musk @elonmusk]

Super Intelligent Beauty

[BRICS News @BRICSinfo]

🇺🇸🇺🇦 Elon Musk says Ukrainian President Zelensky has "gone too far."

[Kalshi @Kalshi]

JUST IN: Zelensky says he will hold elections in Ukraine if Putin agrees to a 60-day ceasefire

[Dawe @Dejvyq]

Most AI agent security is a prompt telling the model to behave
That is not a security boundary. The model is the thing you are trying to constrain. Real boundary: a policy check at the tool call, a credential that expires in 60 seconds, and an audit line for every write

[Dawe @Dejvyq]

Batch writing done for next week. NOMO is coming together, but GitHub billing is still the #1 boss

[Dawe @Dejvyq]

If you're giving an agent an API key, you're giving it the keys to your house. NOMO stops that with scoped IAM