The AI-Pilled Advantage

At the moment, being AI-pilled is the biggest competitive advantage a founder can have. Someone could be 2x better at making decisions about their startup but if you're competing against someone that can do 10x more, you lose almost every time.

The magnitude of this advantage is even larger in SEA than it is in SF. In SF, it's quickly become table stakes for founders but the adoption is much slower in SEA. That's bad for the community as a whole but fantastic if you're one of the few that's taking advantage.

When do you become AI-pilled? There's currently 5 levels of AI sophistication and, in my experience, it happens at Level 3.

1. Basic – You use an LLM (ChatGPT, Claude, etc.) as mostly a Google replacement. You ask for a bolognese recipe, what time it is in San Francisco, etc.
2. Context – You've picked a specific tool and are starting to give it context in the form of projects or memory. You're thinking in terms of a system rather than individual, single prompts.
3. Agentic Execution – You start using LLMs to do things for you, not just answer questions. Often it starts with automating simple workflows (read my e-mail and summarize) but eventually, it works on almost anything.
4. Agentic Engineering – You've automated simple workflows and now that you've had a taste of what they can do, you're drunk with power. You want to automate everything. Anything you can't automate (yet), you want to build something that augments yourself so you can do it faster and better. To do this, you need to start developing your agent harness. You give it more tool access (CLIs, MCPs, etc.), a memory and context management system (mempalace, Kaparthy LLM wiki, etc.), task orchestration (Linear), guardrails and oversight, etc.
5. Agentic Employees – So far, all of the above is used by a single person. For your startup to get maximum advantage, you need agents that work with your organization. They sit in Slack and have their own e-mail. They have their own tasks assigned to them, they write updates and take feedback. For all intents and purposes, they are agentic employees.

To give you a sense of how much we believe in this at Iterative, we're actively working on ways to evaluate people on which level they are (application only sophisticated agents can apply to?) and how to prioritize for them in our investment process.

I'll have to write another post on my experience going through these levels since December (I'm working on 5 now) but if we've been quiet, it's because we've been having passionate internal debates on how Iterative should change given these changes.