Personal Journey

One founder described seeing an AI model generate mobile games with surprising competence. The immediate reaction was a loss of motivation — if the tool can produce what you're building, what's the point?

The reframe: AI is a production accelerator. It compresses the time to get a first version built. It does not compress the time to understand a market, build trust with users, or figure out what actually solves the problem.

What AI Makes Easy (Not Defensible)

Writing code. Generating first versions. Building interfaces. Producing content at scale. If your entire competitive advantage is "I can build this," that advantage is eroding.

What Remains Hard (Defensible)

A Practical Test

Before deciding to build, pivot, or abandon: "If someone used AI to build a competing version of this in a weekend, what would they still be missing?" If the answer is "nothing," you have a problem. If the answer involves relationships, data, domain understanding, or distribution — build there.

Treat the Job Search Like a Market

Don't approach it as submitting applications and hoping. Approach it the way a founder approaches a market: identify the target, research the landscape, build a positioning strategy, and work the channels that give you the highest-leverage access.

Map the ecosystem, not just the target. If your goal is to work at a company like Anthropic or OpenAI, look at the startups they've invested in or partnered with. Those companies are smaller, hiring aggressively, and less likely to have a rigid HR pipeline. A senior role at a partner company can be a faster path than applying through the front door.

Leverage the networks you already have. MIT alumni networks, founder communities, WhatsApp groups — these exist and most people underuse them.

How you enter determines your level. Coming in through a referral from a senior contact positions you differently than applying cold. The decision about what level to offer someone is more arbitrary than most people realize — how you arrive is a meaningful input.

The Delegation Skill

The hardest part of delegation is not the mechanics. It's deciding to let go. Accepting that someone else may not do it the way you would, and that this is fine. The best founders are defined by their ability to find people who are better than them at specific tasks and give those people room to work. This is also one of the things VCs evaluate most closely.

One founder described feeding his pitch deck and background materials into Claude and then using it to role-play investor questions before a meeting with Yale Endowment. Many of the AI-generated questions matched what Yale actually asked.

This is a concrete, low-effort, high-value use of AI that applies broadly: interview prep, pitch practice, objection handling, scenario planning. The model is good at generating the questions a sophisticated counterparty would ask. It can't replace the human judgment needed to answer those questions well, but it can make sure you're not caught off guard.

One founder was weighing mobile gaming (where AI is lowering barriers rapidly) versus robotics (where the market is less explored).

The group's framing: This is less about picking the "right" industry and more about picking the right angle within whatever industry you choose.

If you're competing on the ability to produce software, you're in a race where the finish line keeps moving closer to the start. If you're competing on domain understanding, customer relationships, or creative vision — you're running a different race with fewer competitors.

Rather than choosing between fields, consider combinations. The intersections between fields are often less crowded than the fields themselves.

One reference point from the discussion: David Merrill, an MIT Media Lab PhD who started with tactile gaming cubes (Siftio), then pivoted into robotic delivery drones. The thread connecting the two wasn't the product category — it was a consistent focus on physical human-machine interaction. Finding your own connecting thread is more productive than picking a lane based on which one feels less threatened by AI.

A participant working at Netflix shared an observation that landed with the group: after the initial jump in productivity from AI-generated code, the rest of the software development lifecycle hasn't caught up. Organizations are accumulating fast first versions of things, followed by a long tail of tech debt in productionizing them — compliance, edge cases, alignment.

This maps to a broader pattern. AI makes starting easy. Finishing is still hard. The compliance work (COPPA, privacy regulations), the data model changes, the testing, the thousand small details that separate a prototype from a product — that work is often more time-consuming than the initial build, and AI handles it less reliably.

For founders building software products: the first 80% gets faster. The last 20% doesn't, and that last 20% is often where the actual quality lives.