The most dangerous slide in a Southeast Asian startup deck is not the TAM slide.

It is the U.S. pilot slide.

Every founder expanding into America wants that one recognizable logo. A U.S. enterprise agrees to test the product. Someone senior sounds excited. The company runs a proof of concept. The logo goes into the fundraising deck. Suddenly everyone starts calling it “U.S. traction.”

I would be careful.

A pilot is not traction. A pilot is a question.

The question is not only whether the product works. That is usually the easy part. The harder question is whether the customer has a real problem, a real buyer, a real budget and a real rollout path.

I call this the Four Reals.

Real pain. Real buyer. Real budget. Real rollout.

Without those four things, a pilot is not a bridge into the U.S. market. It is a waiting room with a famous logo on the door.

I have seen many Southeast Asian founders overvalue U.S. interest because the signal feels powerful. A meeting with an American enterprise feels different. A trial with a global company feels like validation. A logo from Silicon Valley, New York or a Fortune 500 company changes how investors listen. It gives the team confidence. It makes the company feel global.

But the U.S. market does not reward interest. It rewards conversion.

American enterprises are very good at experimentation. They will take calls. They will run pilots. They will ask startups to prove value. They will introduce innovation teams, digital transformation teams, strategy teams and product teams. Sometimes that is the beginning of a real commercial relationship. Other times, it is just the enterprise outsourcing its curiosity to startups.

That is the trap.

The founder thinks the company is entering the U.S. market. The enterprise thinks it is learning.

For AI startups, the trap is even worse. Every company wants to experiment with AI. Very few know exactly how to deploy it, govern it, integrate it and pay for it at scale. A demo can be impressive and still fail to become workflow. The model can be strong and still get stuck because the data is messy. The user can love it and still lose to compliance. The champion can be excited and still fail to get finance approval.

This is why I would rather see one boring paid U.S. customer than five glamorous pilots.

A paid customer tells me someone has crossed the internal line from curiosity to commitment. A pilot only tells me someone was interested enough to try.

For founders, the discipline should begin before the pilot starts. Define success in writing. Tie the pilot to a business metric, not a vague feeling. Identify the economic buyer, not just the friendly user. Understand who signs the purchase order. Ask what budget this comes from. Map the security, legal and procurement path early. Set a clear end date. Most importantly, agree upfront on what happens if the pilot works.

A good pilot should create a buying decision.

A bad pilot creates activity.

This matters for Southeast Asian founders because capital efficiency can hide go-to-market weakness. Many founders from this region are excellent at building with fewer resources. They are used to fragmented markets, different cultures and difficult operating environments. That is a strength. But the U.S. is difficult in a different way. It is not fragmented by geography as much as it is fragmented inside the customer.

The user is not always the buyer. The buyer is not always the budget owner. The budget owner is not always the decision maker. The decision maker may still need legal, compliance, procurement, finance and IT to say yes.

So when a founder tells me they have a U.S. pilot, I do not ask whether the logo is impressive.

I ask what happens next.

Is the pain real? Is the buyer real? Is the budget real? Is the rollout real?

If yes, that pilot may be the start of U.S. traction.

If no, it may only be a very expensive conversation.

For twenty years, software investors underwrote a simple assumption: software scales better than labor. SaaS sold access to reusable code. Add another customer, seat, or department, and the marginal cost was close to zero. That was the economic magic behind 80–90% gross margins, high net retention, and the valuation framework that built modern enterprise software.

AI agents complicate that assumption. They are software, but they behave economically like labor. Every completed task carries a cost: inference, tool calls, orchestration, retrieval, verification, retries, exception handling, and sometimes human escalation. In traditional SaaS, usage was mostly evidence of retention. In agentic AI, usage is also a cost event.

The obvious story is that AI agents will replace seats. Why pay for fifty users when one person with agents can do the same work? That story is directionally right, but it misses the harder question. Replacing seats does not automatically create a better business model. A seat is predictable revenue. Work is variable cost.

That is the agent margin trap. Outcomes-based pricing sounds elegant: pay per resolved ticket, qualified lead, invoice processed, or claim reviewed. The customer only pays when value is delivered. But alignment is not the same as margin. An agentic AI company that sells more outcomes also takes on more work, more model calls, more edge cases, more failed attempts, and more dispute risk over whether the outcome was actually achieved.

The bull case is that inference costs will fall. They will. But inference is only the visible cost of autonomy. The hidden costs are verification, exception handling, trust, and liability. Those do not automatically collapse with GPU prices. In many workflows, they become the real cost center.

This is why the key diligence question for agentic AI companies is changing. It is not how much ARR they have, how many workflows they automate, or how accurate the model is. The real question is: what is the contribution margin per outcome after inference, tools, orchestration, verification, exceptions, and human escalation?

If that number does not improve with scale, the company is not AI-native SaaS. It is tech-enabled services with a better interface.

The best agentic AI companies will operate in domains where the work is frequent, narrow, measurable, repeatable, and low-liability. Customer support resolution, invoice processing, compliance documentation, sales research, data enrichment, claims triage, and recruiting screening are good examples. In these markets, success can be defined, failure can be measured, exceptions can be reduced, and margins can improve.

The weakest agentic AI companies will operate where the work is ambiguous, judgment-heavy, high-liability, and hard to verify. Strategy, complex legal advice, medical decisioning, financial planning, enterprise sales, and executive recruiting may demo well, but scale poorly. The edge cases become the product. The exception queue becomes the company.

Investors are trying to value AI labor with SaaS multiples. But labor and software are different economic objects. Software scales by replication. Labor scales by execution. SaaS monetized access. Agents monetize work.

The next great AI companies will not be the ones that automate the most work. They will be the ones that automate the most profitable work.

AI does not kill SaaS. It forces us to ask a better question: is this software, or is this labor with an API?

The $330B number is real. It just doesn't belong to you. Build accordingly.

Samir Kaji recently reignited an important conversation about venture capital. A follow-up piece in Venture Notes, titled “The VC Playbook Has Changed… But Not Equally for Everyone,” argues that while AI has expanded the ceiling of outcomes, the structural advantages remain concentrated among Tier 1 funds with privileged access.

The logic is compelling. Venture capital is still governed by power laws. A small number of companies will generate the overwhelming majority of returns. AI may produce larger outcomes than previous cycles, but unless a fund consistently gains access to those rare companies, the traditional venture math still applies. In that world, brand, proximity, and elite networks continue to matter most.

I agree with much of this analysis. But I believe the more important question is not whether the ceiling has expanded. It is whether the moat has changed.

For decades, access has been the defining moat in venture capital. Proximity to Stanford or MIT. Embeddedness in founder networks. Brand gravity that attracts the strongest entrepreneurs. Access is fundamentally a social moat.

But AI is quietly reshaping another dimension of the industry: detection. And detection is a computational moat.

Much of the current debate focuses on magnitude: larger markets, bigger companies, higher valuations. Yet the more structural shift may be time compression. AI startups differ from prior generations in three key ways. They are more capital efficient. They reach meaningful revenue more quickly. And they distribute globally from day one.

If exit timelines compress from twelve years to six, the economics of venture capital change dramatically. A 10x return over twelve years implies roughly a 21 percent internal rate of return. The same 10x outcome over six years implies closer to 47 percent IRR. That difference alone reshapes fund construction.

Under compressed cycles, you no longer need a single 100x outlier to drive performance. A portfolio of disciplined 8–15x outcomes within shorter timelines can generate exceptional returns. The power law does not disappear, but it becomes denser. The distribution thickens in the middle.

This is where an AI-native fund becomes structurally different.

Most venture firms remain human-limited systems. Sourcing depends on warm introductions and inbound flow. Screening relies on partner memory and qualitative judgment. Portfolio construction follows long-standing heuristics. An AI-native fund instead treats sourcing and scoring as continuous, probabilistic processes.

Rather than waiting for founders to pitch, it monitors real-time signals: GitHub velocity, hiring graph expansion, API usage growth, enterprise traction, technical co-founder networks, and semantic similarity to historical breakout companies. Discovery lag collapses.

Earlier detection leads to lower entry prices, stronger ownership positions, and faster DPI timing. That is not incremental operational improvement; it is structural alpha.

The Venture Notes argument assumes access remains scarce and durable. That may hold for frontier AI labs. But across applied AI, vertical SaaS, infrastructure, robotics, and AI-enabled hardware, information asymmetry is shrinking. High-signal founders leave data exhaust long before demo day. Technical velocity surfaces in public repositories. Community adoption is measurable in real time.

The next Tier 1 fund may not be the most connected. It may be the most computational.

This does not eliminate human judgment. Venture remains an art. But capital allocation can become partially algorithmic without losing its qualitative core. Portfolio construction can move from rules of thumb toward probabilistic optimization: precision at the top of ranked opportunities, founder quality thresholds, regime detection across market cycles, and dynamic reserve allocation informed by updated signals.

The deeper debate, then, is not simply whether the venture playbook has changed. It is whether information asymmetry is still defensible.

If access remains the primary moat, the existing hierarchy persists. AI tools become incremental improvements layered onto a social network model. But if detection becomes the decisive moat, the hierarchy shifts. Structural advantage migrates from proximity and brand toward data and computation.

Access is a social moat. Detection is a computational moat.

An AI-native fund is a deliberate bet on the latter. And in a world defined by compressed cycles and accelerating signal formation, that bet may matter more than legacy status.