Last week at BlueDot, we ran something between an intensive course and an incubator. Seven participants spent five days in our office working through threat models, exploring interventions, and developing concrete project ideas for making AI go well.

We set ourselves a clear (and pretty ambitious) goal on Day 1: find at least one co-founder pair ready to build something that helps make AI go well. We're still figuring out if we hit that goal, but we learned a ton along the way.

## What we shipped

By Friday, we had six problem-solution documents covering everything from datacenter security to hidden model guardrails to AI-enhanced crisis simulations. We also developed [a builder template](https://docs.google.com/document/d/1ziUjrt_8rWLmScwNQdprwDuDmNRWHt6tmTZnlAhGRzM/edit?usp=sharing) that takes people from kill-chain analysis to actionable execution plans — something we'll definitely use again.

And, the wins were real: One participant went from an initial thought on Tuesday to a compelling pitch by Friday. Two more came in with more of _let's sell AI agents_ mindset and left with a plan to _stop sleeper agents_. Everyone shifted to more product-focused mindsets.

## What we learned

**1. We adjusted the format as we went.** We'd originally invited participants to an enhanced in-person AGI Strategy Course cohort, but on Day 1 we realized we were going to run something more like an incubator. We communicated this shift to the group and by Wednesday we'd fully moved into deep problem exploration, pair co-working sessions, and pitch development. This adaptive approach worked well for learning, but it also meant we missed some structure — no culminating pitch day, not enough time blocked for execution as the template came together, and some whiplash for participants who'd signed up for a course.

**2. Co-founder matching is really hard.** By day one it was clear to us that none of our participants were going to start something together. This wasn't a failure of individual talent — everyone was sharp — but we didn't do more work upfront to select for co-founder fit. Going forward: better screening (interviews, writing samples), clearer positioning about what this actually is, and doing more to attract people who already come as pairs (friends, class mates, even co-founders…).

**3. Different problem types need different tracks.** We had technical people, policy folks, and generalists all in one room. Great for diverse perspectives, but this made it harder to go deep on execution. Next time we're extending our track split to the incubator: policy entrepreneurship, research entrepreneurship, and venture entrepreneurship. Same problem exploration phase, but a more constrained solution space. If you have takes on which one we should focus on, please reach out.

**4. In-person makes a real difference.** The exchange of ideas, development of plans and possibility of unexpected collaboration are essential for founding good organisations, and all of these are significantly enhanced by being in the same room as other talented, like-minded people. This is key to what we at BlueDot are trying to do.

## So what's next

We're running version 2.0 in November (Nov 17 - 21). Here's what we're changing:

- We'll choose a track from our list above and run a more comprehensive application process
- Collapse the "when will AGI arrive / be good" debate into day one
- Days 2-3: Deep problem exploration and kill-chain development
- Days 4-5: [Template completion](https://docs.google.com/document/d/1ziUjrt_8rWLmScwNQdprwDuDmNRWHt6tmTZnlAhGRzM/edit?usp=sharing) and pitch day with external experts

The [AGI Strategy Course](https://bluedot.org/courses/agi-strategy) (Round 2 starts October 27…) will remain our primary feeder. We'll filter the best participants through problem statement quality, then interview before inviting folks to the incubator week.

## The bigger picture

There's more talent interested in AI safety than existing organizations can absorb. And there are more risks to mitigate and defenses to build than current organizations can manage. It’s obvious to us that we need to create new organizations, not just funnel people into existing ones, if we are to make AI go well. This intensive/incubator was our first attempt to do that — to take people who are both capable and mission-aligned, give them the structure to go from problem to solution, and back them to build.

Last week was messy and energizing in equal measure. We're learning fast and are only getting started. If you're interested in joining a future incubator week, [apply to the AGI Strategy Course here](https://apply.bluedot.org/agi-strategy-course) and we'll be in touch.


What we learned from turning our AGI Strategy Course into an Incubator

Josh Landes, Aniket Chakravorty • October 13, 2025