A common question I get from founders right now is some version of:
“We have traction. Why is it still so hard to raise?”
It’s a fair question.
A lot of companies have users, revenue, strong demos, and fast product velocity. Yet the investor response is often vague. “We’re being more selective.” “We’re looking for deeper technology.” “We’re focused on harder industries.” “We want to see more durability.”
Some of that is real. Some of it reflects genuine uncertainty in how this market should be underwritten.
I feel this in my own conversations too. There are companies I would have known how to categorize a few years ago that are harder to read now.
The honest answer is that we are still learning how to underwrite this category. AI has made early traction easier to generate and harder to interpret. It has made products faster to build, and many companies look promising before it’s clear whether they are durable.
That’s the tension.
Founders often assume traction should clarify the company. In AI, traction can be real and still not prove what investors need to believe.
So the useful question is not simply: does this company have traction?
It is: what does the traction actually prove?
The hard part is not only generating traction. It’s knowing whether that traction reduces the risks that matter.
The way I’ve come to think about it is this:
AI doesn’t change what startups are. It changes where they break.
If you understand where companies are likely to break, you get a clearer sense of what traction matters, what risks remain, and ultimately where value accrues.
Most AI writing asks, “What can AI do now?” I think the better question is, “What does that change about how companies get built?”
The chain is fairly simple.
If capability becomes abundant, then building becomes less scarce.
If building becomes less scarce, then early traction becomes easier to generate and harder to trust.
If software starts doing work rather than simply assisting with it, then the bottleneck moves from product quality to deployment, trust, and distribution.
And if that is where companies break, then value accrues to the teams that can own workflows, earn trust, and reach the buyer.
That’s the shift underneath most of what we are seeing right now.
Why the old signals are noisier now
A lot of AI companies are still being evaluated with a familiar lens. Is the product better? Is the UX cleaner? Is growth fast?
Those things still matter. But they’re no longer where the main constraint sits.
Capability is no longer scarce. Strong models are increasingly available to everyone, and small teams can ship products that would have taken much larger teams not long ago.
Which leads to a slightly uncomfortable observation:
You can build something that looks like a real company before it is clear whether one is forming.
That shows up more often in AI than in previous cycles.
Usage has always been an imperfect signal. But in AI, it is easier to misread because curiosity, novelty, demo value, and cheap generation can all look like progress before workflow ownership is proven.
The economics can distort the signal too. If customers are adopting an AI product because the underlying usage is unusually cheap today, founders need to understand whether demand holds when pricing, usage limits, or inference costs normalize.
What investors are really underwriting
This is why, in this environment, the strongest founders are not the ones who pretend the market is obvious.
They are the ones who can explain how they’re thinking through it.
They can say what their traction proves and what it doesn’t prove yet. They can explain why they chose a specific wedge, what they are learning from customers, where workflow ownership might emerge, and what risks still need to be retired.
That depth of thought matters because in a fast-moving market, investors are not only underwriting the current metrics. They’re underwriting the founder’s ability to navigate ambiguity as the market takes shape.
Early traction used to answer more of the question. Now, it often raises the next set of questions.
Put differently, investors are not always asking for more traction. They are asking for better evidence: evidence that the product can get into production, stay there, expand from there, and reach the buyer repeatedly.
That does not mean traction matters less. It means founders need to explain what kind of evidence it is.
The bottleneck moves
At a high level, AI reduces the cost of doing certain types of work. As products move from answering questions to taking actions, software starts to do the work rather than simply assist with it.
But once software starts to act, a different set of constraints shows up.
Execution introduces risk. Mistakes matter more. Edge cases matter more. Accountability matters more.
Organizations respond in a fairly predictable way. They slow things down, add controls, and become more cautious about what actually gets deployed.
This is where the bottleneck begins to move.
It is not capability that limits adoption. It is whether something can be deployed safely into a real workflow.
One thing that is easy to miss is how much the timeline has compressed. In previous cycles, it could take years to see where a company’s weaknesses would show up. Now, products reach real usage quickly, and failure modes show up faster. The feedback loop between building, deploying, learning, scaling, and failure is much tighter.
That compression changes how founders build and how investors evaluate risk.
It also makes the market harder to read. Companies can look mature before they’re durable. Categories can look crowded before buyers have decided what the category really is. And adoption can be uneven because different customers are willing to trust different levels of automation.
The result is a messy middle: many players, overlapping products, uneven adoption, and no clear winner early on.
But that does not mean value stays fragmented forever. Once a company earns trust, owns the workflow, and reaches the buyer, adoption can accelerate quickly.
Where the durable value forms
In the last software cycle, a lot of value accrued to systems of record: the places where teams stored customer data, financial data, employee data, or operational truth.
In this cycle, the more interesting question is which companies become systems of work: the layer where tasks are executed, exceptions are handled, and decisions turn into action.
That’s why workflow ownership matters.
The closer a product gets to real execution, the more trust it needs. The more trust it earns, the more value it can capture.
A product that answers questions can be useful. A product that owns a workflow can become embedded. A product that is trusted to act can become infrastructure.
That is also where defensibility starts to form: not because the model is unique, but because the company has earned workflow depth, trust, and distribution that are difficult to replicate.
That is the progression that matters.
Where founders can get stuck
What this looks like in practice is fairly consistent.
Founders start too broad, often trying to define a platform before they have a clear entry point. They optimize for demos because demos convert. They use early usage as a proxy for value, even when it does not translate into durable workflow ownership. Trust and controls get pushed to a later phase. And there is a tendency to wait for the market to settle before committing to a specific wedge.
None of these are new. They just show up faster now, and some of the signals are easier to misread.
The product that works in a demo but cannot pass a security review may not be deployable yet.
If you do not own the workflow, it is difficult to own the outcome.
And if you do not have a path to distribution, even a trusted product can remain niche.
What this implies in practice
If the bottleneck has shifted, the strategy shifts with it.
It tends to work better to start with a narrow, clearly defined workflow and focus on executing that workflow end to end. From there, the priority is making the system deployable in a real environment, which usually means thinking about permissions, auditability, failure modes, and distribution much earlier than teams expect.
Expansion comes after that, once the system is trusted to do real work.
The companies that take this approach often do not look the most ambitious at the beginning. But they tend to build something more durable over time because they sit inside the workflow rather than next to it.
Closing
Every cycle has a moment where something becomes much easier. That usually shifts the constraint somewhere else.
Right now, building has become easier.
So the constraint has moved to real-world execution, trust, and distribution.
I don’t think this cycle changes the fundamentals of building companies. But it does change where the hard parts are.
For founders, that means traction is not the end of the story. It is evidence that needs to be explained.
If you can show what your traction proves, what risks it removes, and where the company is built to hold under pressure, you are probably pointed in the right direction.