What I Heard in a Room Full of Builders and Funders

What I heard in a room full of builders and funders

My professional home has largely been in institutions: universities, philanthropy, international development, organizations that think in decades and sometimes take years to make consequential change. I have, however, also worked with high-growth company founders, and many philanthropists also happen to be investors.

Which may be precisely why I’m curious about startups.

A century-old institution and a three-month-old company can seem like opposite species. They operate at different speeds, with different relationships to capital, risk, decision-making and change. But both are continually making choices about futures that don’t yet exist: what to build, what to fund, what to believe, and what becomes possible next.

I’m interested in what happens when ideas travel between those worlds. Sometimes the fastest way to see your own world differently is to spend time in someone else’s.

So I recently found myself in a room in Vancouver with founders, funders, builders and people working around the machinery that supports them, listening to a conversation about how AI is changing not only how companies are funded, but how they are built.

I went mostly because I was curious. I left with a notebook full of things I haven’t stopped thinking about.

Building is getting easier. That changes what building proves.

Not long ago, an early-stage founder might approach an investor seeking capital to build a first prototype. Increasingly, the expectation is that the prototype already exists.

I heard of investors going further: seeing a startup demo and attempting to reproduce it themselves using AI-assisted development. If they can build something comparable relatively easily, what exactly has the demo proved? If the demo is fairly easy to reproduce with AI by someone who is not a software developer, that has material implications on the investor’s decision on whether to fund the company or not.

The product may still be useful. The accomplishment may still be real. But the existence of the thing tells us less about whether the company possesses an advantage that will last.

Domain expertise may matter more, not less.

Another recurring idea concerned the relationship between technical capability and deep knowledge of a particular field.

Historically, bridging those worlds could be difficult. Someone who understood biology, finance, manufacturing or another specialized domain might depend heavily on technical specialists to translate that knowledge into software.

AI changes some of that equation. If implementation becomes more accessible, people who deeply understand the underlying problem can participate more directly in building the solution.

That doesn’t make technical expertise irrelevant. But it may move the bottleneck.

• Knowing how to make something is valuable. Knowing what actually needs to be made, why, and for whom may become more so.

The bar to build has fallen. The bar to sell has not.

This was one of the simplest observations of the evening, and one of the insights that stayed with me.

More people can build credible products. That does not mean more people can persuade someone to buy them.

Sales, relationships, community and network may therefore become more valuable rather than less. And another capability surfaced alongside them: clarity of direction. Critical thinking.

Large language models are remarkably capable at producing and consuming information. But someone still has to decide what that information is for. What matters? What should remain consistent? What story should it ultimately tell—to a customer, investor, user or team?

• When generation becomes abundant, the ability to generate may become less interesting.

The ability to direct generation may be another matter.

An impressive AI demo and a dependable AI system are not the same thing.

Another builder described a pattern he sees inside organizations adopting AI. A team identifies a workflow, builds a prototype, sees that it works—and gets excited.

Then comes the much harder task of making it reliable enough to become part of the actual operation.

AI systems are probabilistic. Edge cases appear. Outputs vary. Context changes. Something that performs beautifully in a demonstration may behave differently when it encounters the messiness of real organizational life.

One approach discussed was to keep people inside the loop while the system earns greater autonomy: observe how AI is actually being used, test reliability and gradually reduce human involvement where the evidence supports it.

Capability is not the same thing as dependability. And a prototype is not an operating model.

Smaller teams don’t necessarily make human capability less important.

AI is already beginning to change organizational structures. Roles can consolidate. Some forms of knowledge work can be automated. Companies may be able to reach milestones with substantially smaller teams.

But one builder framed that leverage in a way I found more interesting than the usual discussion of jobs eliminated or hours saved: AI should allow people to spend more of their capacity on higher-impact decisions requiring accountability and intuition.

That shifts the question from What work can AI take away from people? to What work becomes more important for people to do?

I’m increasingly interested in that second question.

What an organization remembers may matter more too.

One of the evening’s more unexpected threads began with research-and-development tax credits.

If AI produces more of the code, where did the meaningful human work occur? Perhaps it was in identifying the uncertainty, designing the experiment, choosing between approaches, interpreting failure or deciding what to try next.

That makes the record of thinking more interesting than a simple record of output.

The conversation expanded from there into documentation, intellectual property and organizational knowledge. One particularly useful observation was that AI can multiply the quality of the knowledge available to it—including poor or outdated knowledge.

Which made me wonder what happens when organizations begin giving increasingly capable systems access to memories they have never organized particularly well.

And then there is the problem of consensus.

Near the end came one of the ideas I found hardest to shake.

Large language models are extraordinarily good at working with patterns derived from what already exists. But exceptional companies are often built around something that is not yet consensus: an overlooked problem, unexpected behaviour, a market others have misunderstood, a technology whose significance others haven’t recognized.

That doesn’t make consensus bad, nor contrarianism inherently intelligent. But it raises an intriguing question: Where does the idea that doesn’t yet make sense come from?

I suspect I’ll be thinking about that one for a while.

I went in thinking about AI. I came out thinking about scarcity.

None of these observations, on their own, offers a complete theory of what AI will do to companies. Some may prove durable. Some may look quaint surprisingly quickly. We are trying to interpret a shift while standing inside it.

But some stubborn, intriguing questions began to emerge between them.

• When making becomes easier, what becomes difficult?

• When information becomes abundant, what becomes scarce?

• When competent execution becomes widely available, what distinguishes one organization from another?

• When machines can participate in more of the work, what happens to expertise, judgment and memory?

I came into the room curious about how AI was changing the economics of building companies. I left thinking about scarcity, expertise, trust, memory, judgment—and what remains difficult when making things becomes easier.

Some of those questions belong to startups.

• I’m not convinced they belong only to startups.

::::::::::::::::::::::::::::::::::::::::

Contemplations & Provocations | A field essay by
Marie Fe Isla Rae

© 2026 Marie Fe del Rosario. All rights reserved.
Terms of Use · Privacy Policy · Permissions & Collaborations

All content on this site—including text, frameworks, methodologies, visuals, and strategic concepts—is protected by copyright and may not be reproduced, adapted, or used without prior written permission.

Authorship and web design by Marie Fe.

Field notes on consequential moments, creative strategy, experience, narrative, design—and the work of making possible futures tangible.

For people interested in what happens between imagining and making.