I have been thinking about the difference between the surface and the interface in AI, and I increasingly think the surface may matter far more than the interface.

Instinct on WhatsApp is a good example. Instinct can create a really good AI experience where you send a voice note, tell it what you want done, and it goes and gets that work done for you. But Instinct is essentially the interface. WhatsApp is the surface. And I think that distinction matters quite a bit.

The interface can get better very quickly, but it can also get copied and commoditised very quickly. Models improve, features get replicated, and things that feel differentiated today can become fairly common over time. The surface is much harder to replace because it is already deeply embedded in people's lives.

If you look at WhatsApp, it already has your conversations, relationships, habits and years of usage sitting inside it. iMessage has a similar advantage because of how deeply it is embedded into the iPhone and into how people communicate every day. So even if these surfaces are behind on some of the new AI experiences initially, they can catch up and still remain extremely powerful because the user is already there.

In some ways, Instinct shows how well this can work. You can put a completely new AI experience inside an existing surface and suddenly enable a lot more without asking the user to change behavior very much. You send a voice note, which is something you already know how to do, except now that voice note can actually result in work getting done.

But there is another part of this that I find even more interesting. You could argue that AI also gives us an opportunity to build a completely new surface.

Most of the computing surfaces we use today were built around the way we had to interact with computers. We needed screens, keyboards, apps, browsers, menus and touchscreens because that was the setup required to tell software what to do. In a way, we learned how to operate computers because computers could not really understand us.

AI changes that quite dramatically. If intelligence can understand language, voice, context and intent, then increasingly I should just be able to tell the computer what I want done and have it figure out a lot of the rest. I do not necessarily need to sit in front of a screen, open the right application, navigate through the workflow and then do everything myself.

And I think that creates an opportunity to build a new surface for an intelligence led, agent first world.

What that surface looks like is much less obvious. It could still be the phone. It could be glasses. It could be an earbud, a ring, a bracelet, or something completely different. We probably do not know yet, and I am not sure the form factor itself is even the most important part.

The more interesting question is whether we can build a surface where you are no longer constantly operating the computer. You express what you want, and intelligence does more and more of the work around you.

I think both things can happen at the same time. Existing surfaces like WhatsApp and iMessage are going to become much more powerful as AI gets built into them, and their existing distribution, habits and user love give them a huge advantage. But AI may also give us an opportunity to rethink the surface itself and build something that is native to this new way of interacting with computers.

Interfaces can get copied and commoditised.

Surfaces are much harder to displace.

And that is why I increasingly think the surface may be where the more durable power sits.

Incumbents have a distribution advantage.

Startups have a concentrated ambition advantage.

As AI models become broadly available, the conventional view is that distribution becomes more important. If everyone has access to similar intelligence, then the companies with hundreds of millions of users, capital, data, infrastructure and customer relationships should be in the strongest position.

That is true.

But it may still miss the more important shift.

For a large company, any new AI opportunity is one among many. It competes with existing products, revenue streams, internal priorities and organizational attention.

For a founder, it can be the entire company.

Every person, every dollar and every decision can be pointed at one market and one outcome.

That is concentrated ambition.

This has always been a startup advantage. What is new is the execution power now sitting behind it.

Historically, ambition was constrained by how much a small team could actually build. Large products required large teams, long development cycles, significant capital and time. Even the most driven founder could only move so fast.

AI changes that.

A small, exceptional team can now build, test, iterate and ship at a speed that would have required a much larger organization only a few years ago.

AI is making concentrated ambition far more powerful.

That creates a force that did not exist before: founder-level focus combined with an execution speed that can meaningfully offset the incumbent's advantages in distribution, capital and scale.

This is why the assumption that incumbents will naturally win as models commoditize may be incomplete.

Distribution still matters enormously.

But concentrated ambition, turbocharged by AI-led execution, now matters much more than it used to.

I think the infinite software era is here.

For most of the history of software, the constraint was pretty obvious. Software was expensive to build. Engineering teams were expensive. Every new product created more code to maintain, more infrastructure, more integrations, more edge cases and more things that could break.

So companies had to choose.

You picked a problem, built a product around it and became very good at that product. If you were building software for freight forwarders, maybe you did shipment visibility. Someone else did bookings. Someone else built the ERP. Someone else built the CRM. Someone else handled payments, accounting or financing.

Entire software categories emerged because there simply wasn't enough software to go around.

I think AI changes this much more fundamentally than we currently appreciate.

It isn't simply going to help companies write the same amount of software with fewer engineers.

Companies are going to write vastly more software.

And eventually, the best companies are going to write almost everything their customers need.

That is what I mean by Infinite Software.

From products to complete platforms

Imagine you are building an AI-native company for freight forwarders.

Today you might start with one painful workflow. Perhaps quoting. Perhaps document processing. Perhaps shipment tracking.

But once you have the customer, the data, the integrations and the intelligence required to understand how that freight forwarder actually operates, why would you stop there?

You can do bookings.

Then tracking.

Then customer communication.

Then documentation.

Then invoicing.

Then payments.

Then collections.

Then CRM.

Then accounting.

Then eventually much of the ERP itself.

Historically you could argue that this was a terrible idea. It would require hundreds of engineers, years of development and a giant product organization. The company would lose focus long before it ever got there.

But what happens when an extremely good engineering team has built its own AI development infrastructure, coding agents, testing systems, evaluation harnesses, design systems, deployment infrastructure and internal frameworks that allow it to ship software at 10x or 50x the previous rate?

At some point the question flips.

It is no longer:

Why should we build this?

It becomes:

Why would we not build this?

If the customer already trusts you, the data is already sitting inside your system and the new workflow naturally connects to everything else you provide, the marginal cost of building the next piece of software starts becoming very small relative to the value of owning the workflow.

That is a completely different software world.

Mercury Books is a perfect example

Mercury started as a bank for startups.

Now it has launched Mercury Books, a full double-entry accounting product sitting directly inside Mercury. Transactions, reconciliation, journal entries, P&L, balance sheet, cash flow, AI categorization, all tied directly into the place where the money already moves.

If you think about the old software world, this is kind of remarkable.

Why is a neobank building accounting software?

Because in the infinite software world, that might be the wrong question.

The better question is: why shouldn't it?

Mercury already has the customer. It already has the financial data. It already sees banking, cards, invoices and bill payments. Accounting is an adjacent workflow sitting directly on top of the same data.

And once accounting is there, there will be another adjacent workflow. And another one after that.

Mercury itself describes Books as part of a future where a company's entire financial life can live inside Mercury.

That is the direction I think a lot of software is going.

Not a bank with a few additional features.

A complete financial operating system built around a customer Mercury already owns.

ElevenLabs is another version of the same thing

Look at ElevenLabs.

At one level, ElevenLabs owns foundational voice infrastructure. Developers can build applications on top of its models and APIs.

But ElevenLabs is now also offering Reception, a finished AI receptionist that a small business can turn on in minutes. It answers calls, understands the business, books appointments, transfers calls, takes messages and handles customer interactions.

There are already plenty of startups building AI receptionists.

So should ElevenLabs stay underneath them as infrastructure?

I don't think that question has an obvious answer anymore.

If ElevenLabs already has the models, the voice infrastructure, the agent platform and the underlying technology, why wouldn't it package some of the highest-value applications itself?

And then why stop at reception?

You can imagine the same infrastructure expanding into sales calls, support, collections, scheduling, lead qualification, customer service and dozens of other voice-heavy workflows.

Again, the boundary between infrastructure and application starts becoming much less rigid.

The model company writes applications.

The bank writes accounting software.

The vertical application writes the ERP.

Everyone moves into everyone else's territory.

Software abundance changes company boundaries

I think this is the important part.

We have spent decades thinking about software companies using boundaries that were partly created by the economics of producing software.

CRM was one category.

Accounting was another.

Banking was another.

Customer support was another.

ERP was another.

These boundaries felt natural because companies were built around them.

But many of them may not actually be natural.

They may simply be artifacts of a world where software was expensive to produce.

Once software becomes abundant, those boundaries begin to collapse.

A company with the strongest relationship with a customer can simply keep expanding around that customer.

A company with the strongest underlying infrastructure can keep moving upward into applications.

A company with the best system of record can keep adding workflows around the data it already owns.

And a vertical AI company can keep expanding until it starts looking less like a point solution and more like the operating system for an entire industry.

This is why I think AI-native software companies may look very different from SaaS companies of the last twenty years.

The goal may no longer be to dominate a category.

The goal may be to cover the customer.

The scarce thing will no longer be software

This doesn't mean every company should randomly build hundreds of features.

There is still such a thing as bad software. There are still products customers don't want. There are still workflows where you lack credibility, data or distribution.

But I think the constraint is moving.

The scarce resource used to be the ability to produce software.

Increasingly, the scarce resources will be:

customer ownership,

distribution,

proprietary data,

workflow context,

trust,

and permission to do more.

Once you have those things, software becomes almost infinitely expandable around them.

This has pretty profound competitive implications.

For years we have talked about software moats in terms of features. But features become much less defensible if every good company can manufacture software extremely quickly.

The moat moves somewhere else.

Who owns the customer?

Who owns the data?

Who sits inside the workflow?

Who is trusted enough to take the next action?

Who can distribute the next product essentially for free?

Those questions become much more important.

And it also means that product roadmaps themselves may change.

The traditional startup advice was to focus relentlessly. Do one thing exceptionally well. Don't become a bloated suite.

That advice was correct in a world where every additional product imposed enormous organizational and engineering cost.

I am not sure it will remain equally correct in a world where a 30-person AI-native engineering organization can eventually produce what previously required 300 or 3,000 people.

The best companies will still start narrow. They will still need a wedge.

But the wedge may increasingly just be the entry point.

Once they have earned the right to expand, they will.

Everybody will write everything

This is why I keep coming back to the phrase Infinite Software.

We are entering a period where the amount of software produced is going to explode.

Not because the world suddenly has 100x more software companies.

Because every software company can potentially produce 100x more software.

And as that happens, the distinction between product companies, platforms, infrastructure companies and application companies will get increasingly blurry.

Banks will write accounting software.

Model companies will write applications.

Vertical AI companies will write ERPs.

Infrastructure companies will move up the stack.

Application companies will move sideways across workflows.

Everyone will cover more ground.

The winners, I suspect, will be the companies that understand this early and build themselves for it. Not just by using AI coding tools, but by creating engineering systems where software production itself becomes a massively scalable internal capability.

Because once you can manufacture high-quality software extremely cheaply and extremely quickly, leaving an obvious customer need for somebody else to solve starts making less and less sense.

For most of software history, companies had to decide what software not to build.

We may now be entering the opposite era.

Everybody will build everything they have the right to build.

That is Infinite Software.

Every growing AI co in SF will build an India AI office.
Every PE portco injecting AI will build an India AI office.
Every "Rollup & add AI" co will build an India AI office.
Every enterprise AI leader will build an India AI office.

..because there is only so much AI talent in the west.
..because evals scale will make India office inevitable.
..because fine tuning and data will be best done in India.
..because it is perfect for the PE playbook.
..because it is essential to make AI-led rollups work.
..and more.

At InCommon, we recognised this micro-trend early on and built a completely new GCC 2.0 stack designed with the above in mind - helping global companies build their "India AI Office" with talent and culture at its core, not real estate.

InCommon not only delivers a full-stack experience but has carefully designed several programs like The Better Hack (multi-city AI hackathon), GCC Leaders Breakfast (multi-city quarterly gathering of the smartest AI GCC thinkers & leaders) and more to create the right community for the long term.

If you are considering building your "India AI Office" give InCommon a shout.

Humans used to do the work. Then we built software to help. Then we built SaaS to make that software easier to use. But the goal never changed: get the job done.

We didn’t buy Salesforce because we loved CRMs, we wanted more sales. We didn’t implement Workday for the fun of it, we needed org-wide HR coordination. Every layer of software, from on-prem to cloud, was a means to an end.

But something strange happened along the way: SaaS became the job. Setup, configuration, integrations, dashboards, training… over time, more and more human effort shifted from doing the real work to operating the tool designed to do the work.

AI breaks that cycle.

With AI, we’re not just getting better tools - we’re finally getting closer to what we always wanted: results. Not dashboards. Not notifications. Not workflows. Outcomes.

That’s the core of Outcome-as-a-Service. It’s not AI bolted onto SaaS. It’s AI that replaces SaaS — by doing the job and delivering the result.

You don't need a CRM - you want closed deals.
You don't need HR software - you want onboarding done.
You don’t need analytics tools - you want the insight and the action.

We started investing in this shift in 2022–2023, well before Copilots came along as AI bandaids to SaaS. But the real breakthrough isn’t copilots. It’s contracting for outcomes, not effort. And we believe OaaS will become the defining model for the next generation of software businesses.

This is what every enterprise buyer actually wants:

• Less SaaS sprawl.
• Fewer seats.
• More results.

Outcome-as-a-Service is not the evolution of SaaS — it’s the replacement. And it opens the door to a new generation of builders who think in outputs, not inputs.