I have not posted in months, which in internet time means I have either been building something important, overthinking a draft, or being held hostage by my own Notes app.

In this case, I plead guilty to all three.

But a recent a16z essay, Is Software Losing Its Head?, put language to something that has been sitting in the background of enterprise software for a while: the user interface is becoming less defensible. For twenty years, software vendors sold the screen. The dashboard. The seat. The workflow inside their app.

Now agents can reach the data and take actions through APIs, MCP servers, and command-line tools. The human no longer needs to live inside every product for the work to happen.

That sounds like software is becoming weaker. I think the opposite is happening.

Software is not dying. It is moving. The value is shifting away from the visible interface and into the parts customers do not want to rebuild: the data model, permissions, compliance logic, audit trail, workflow actions, integrations, and domain intelligence.

The head is becoming optional. The body still matters.

And that opens up a new category of software company: headless SaaS.

Not SaaS without value. SaaS without the assumption that the vendor’s interface is where the customer must live.

The shift: from software as a place to software as infrastructure

Traditional SaaS was built around a simple idea: log in, use the vendor’s workflow, pay per seat.

That model worked because the interface was the product. Salesforce was where salespeople worked. Workday was where HR worked. Greenhouse was where recruiters worked. Zendesk was where support teams worked.

But AI changes the buyer’s mental model.

When an agent can update a CRM record, screen a candidate, retrieve a payroll policy, trigger a support workflow, or summarize account history without opening the app, the app’s UI becomes less central. The customer starts asking a different question:

Why should my team adapt to your interface if your underlying capabilities can plug into my own operating system?

That is the real shift.

The customer does not necessarily want to rebuild payroll compliance, recruiting workflows, audit logs, permission systems, or tax logic. But they increasingly want to own the layer where their people and agents actually work. They want their own internal tools, their own AI copilots, their own dashboards, their own operating rhythm.

The vendor’s product becomes less like a restaurant and more like a kitchen supplier. The vendor still provides the hard ingredients: regulatory rules, canonical data, workflow primitives, integrations, safe write-back actions. But the customer decides how the meal is served.

That is headless SaaS.

Why this is happening now

Three forces are converging.

First, AI has made internal software cheaper to build. A company can now prototype workflows, dashboards, automations, and internal tools much faster than before. This does not mean every company can build a secure, compliant, enterprise-grade system from scratch. It does mean the cost of building the “head” of software, the interface, the workflow wrapper, the internal app, has collapsed.

Second, enterprise systems are becoming more programmable. Salesforce is already moving in this direction with Agentforce 360: Slack becomes the working surface, Salesforce data and workflows sit underneath it, and agents can operate across the stack through governed connections. Whether you call that “headless,” “agentic,” or “360,” the signal is the same: the interface is no longer the only place where the software lives. Anthropic’s Model Context Protocol points in the same direction by giving agents a standard way to connect to tools and data. Software is being redesigned so agents can consume it directly.

Third, forward deployed engineering is becoming a market motion. OpenAI, Anthropic, Palantir-style teams, consulting firms, and AI-native vendors are all moving toward embedded deployment. The reason is simple: customers want custom systems, but they need help with the last mile, integrations, process mapping, governance, security, and change management.

Put together, these trends create a new possibility: companies can build custom operating layers without rebuilding the difficult primitives underneath.

That is why the opportunity is not simply “AI will replace SaaS.”

The better thesis is:

AI will weaken generic interfaces, while increasing the value of trusted backend primitives.

What remains defensible

It is tempting to say every SaaS company is vulnerable. That is too easy, and probably wrong.

Some parts of software are very hard to replace.

Compliance is defensible. Payroll is the obvious example. Calculating wages is not the hard part. Handling tax filings, local rules, benefits, worker classifications, audit trails, and jurisdiction-specific changes is the hard part. That is why companies like ADP, Rippling, Deel, Remote, Gusto, Workday, and others still matter. Their moat is not the employee profile page. It is the compliance machinery behind it.

Identity and permissions are defensible. In an agentic world, the question “who is allowed to do what?” becomes more important, not less. If agents can take actions across systems, then role-based access, auditability, approval flows, and delegated authority become critical infrastructure.

Workflow primitives are defensible. It is one thing to show information. It is another thing to safely change production state: move a candidate to the next hiring stage, approve payroll, issue a refund, close a support case, or update a ledger.

Network effects remain defensible. Products like LinkedIn, payment networks, marketplaces, and other multi-party systems are not vulnerable just because their screens can be copied. Their value lives in the network, not the UI.

What becomes weak is the part we have often overvalued: dashboards, dropdowns, click paths, and seat-based access to a database.

The UI is not worthless. It is just no longer the moat by default.

What headless SaaS really means

Headless SaaS is not “a normal SaaS product with an API.”

Most enterprise vendors already have APIs. Many of those APIs are incomplete, difficult to work with, or treated as secondary to the main product. The customer still has to adopt the vendor’s workflow and interface.

Headless SaaS inverts that.

The product is the programmable layer: APIs, MCP servers, workflow primitives, permissions, compliance rules, event logs, and integrations. The vendor may still offer a default UI, but the UI is no longer the center of gravity.

A real headless SaaS company would invest heavily in six things:

The “recipe library” matters more than it sounds.

Customers do not just want raw infrastructure. They want to know how to assemble it. A headless ATS vendor, for example, should not only expose candidate, job, scorecard, and interview APIs. It should provide recipes for “screen inbound applicants,” “schedule interviews,” “generate offer letters,” “run equal-employment reporting,” and “sync hired candidates into HRIS.”

This is where forward deployed engineers become important. Their job is not to create endless bespoke consulting work. Their job is to help customers implement the system, learn from the deployment, and turn those learnings into reusable recipes.

That is how the model scales.

The ATS example: why pay for the room if you only need the rails?

Applicant tracking systems are one of the clearest early categories for headless disruption.

Recruiting teams often pay for seats so people can log into an ATS, move candidates through stages, leave notes, schedule interviews, and generate reports. But much of that workflow is now agent-friendly.

Imagine a mid-sized company with a recruiting team that mostly works in Slack, email, calendar, and an internal hiring dashboard. In a headless world, the ATS vendor provides the underlying recruiting primitives:

• candidate profiles

• job and pipeline data

• interview scheduling hooks

• scorecards

• compliance reporting

• offer-letter workflows

• audit logs

• permissions

• integrations into HRIS and calendar systems

The company builds its own hiring interface on top. Recruiters do not need to live inside the ATS. An agent can summarize candidates, schedule interviews, nudge hiring managers, update pipeline stages, draft rejection emails, and surface decisions where the team already works.

The ATS vendor still earns money. But it earns it from workflow execution, API usage, compliance coverage, or hiring volume, not necessarily from every human who needs to look at a candidate record.

This is compelling because it gives the customer control without forcing them to rebuild the sensitive parts.

They own the head. The vendor owns the rails.

HRIS and payroll: harder to replace, very likely to go headless

HRIS and payroll are different.

A company may happily replace the interface of its HR system, but it is much less likely to rebuild payroll compliance. Payroll involves filings, taxes, benefits, local laws, country-specific rules, payment timing, employee records, and audit exposure.

That makes HRIS and payroll less vulnerable as systems of record, but very attractive as headless infrastructure.

The future HR stack may not be “replace Workday with a chatbot.” That is the lazy version of the thesis.

The stronger version is: keep the compliant backend, but let the customer build the employee operating layer they actually want.

A company might use a headless HRIS or payroll provider for:

• employee records

• onboarding and offboarding actions

• payroll calculations

• tax and compliance rules

• benefits logic

• document generation

• approval flows

• audit trails

• employee status changes

• integrations into finance, IT, and identity systems

Then it builds its own internal people OS on top: an AI assistant for employees, a manager dashboard, a finance planning view, a workforce analytics layer, and custom workflows that match how the company actually operates.

This is not less valuable to customers. It is more valuable.

They get the compliance they cannot afford to mess up, while escaping the one-size-fits-all interface.

Why customers care: the intelligence layer

The most compelling reason customers will care about headless SaaS is not cost.

It is context.

Every company is trying to build a more intelligent operating system. They want an AI layer that can reason across sales, finance, people, support, operations, documents, and customer data. But today, that intelligence layer is trapped behind fragmented SaaS interfaces and brittle integrations.

The customer has to pull data out of one system, transform it, pipe it into another, and hope the permissions survive the journey.

Headless SaaS changes the architecture.

If every major system exposes its data and actions through clean APIs and MCP tools, the customer’s AI layer can sit naturally across the business. It can ask the HR system about headcount, the CRM about pipeline, the accounting system about revenue, the support system about customer issues, and the ATS about hiring velocity.

More importantly, it can act.

That is why the shift is compelling. Customers are not just trying to save on SaaS bills. They are trying to own the context layer where their AI lives.

In the old world, every SaaS product wanted to be the place where work happened.

In the new world, the customer wants work to happen in its own operating layer.

Why this is especially compelling for medium-sized businesses

The Fortune 500 will build a lot of this internally. They have large engineering teams, data teams, procurement power, and the patience to run multi-year transformation programs.

Small companies may continue to use packaged software because it is simple and good enough.

The most interesting market is in the middle.

Medium-sized businesses are large enough to have complex operations, but not large enough to justify massive enterprise software spend or permanent internal platform teams. They feel the pain of generic SaaS earlier than people think. Their processes are becoming unique. Their teams are growing. Their data is scattered. Their software bills are expanding. But they cannot always afford to hire the army required to build everything from scratch.

Headless SaaS is almost perfectly shaped for them.

It lets them rent the hard parts: compliance, infrastructure, integrations, and governed actions. Then they can build the differentiating layer: the workflows, dashboards, agents, and operating system that match how their business actually works.

This is why the category could expand quickly. Medium businesses do not want “more software.” They want their systems to finally fit.

What vendors need to do now

Incumbent vendors should not treat this as a feature request. It is a business model shift.

Salesforce is the useful reference point here. Agentforce 360 and the Slack “agentic OS” push are not just AI feature launches; they show how a large incumbent can try to keep the system of record, the data layer, and the governed workflow layer even as the user experience moves into chat, agents, and other surfaces. That is the playbook other vendors should study.

If customers stop living inside the UI, then vendors need to make the underlying product easier to consume, easier to trust, and easier to compose.

The urgent work is clear.

Expose the product properly. APIs cannot be afterthoughts. MCP servers, webhooks, workflow endpoints, event streams, and documentation should become first-class product surfaces. Salesforce is already trying to make CRM data and actions reachable outside the classic Salesforce UI; that is the standard customers will start expecting everywhere.

Make actions safe. If agents are going to execute workflows, vendors need granular permissions, approval steps, audit trails, rollbacks, and strong observability.

Build recipes, not just endpoints. Customers do not want a thousand API docs and good luck. They want implementation patterns: “how to build a recruiting agent,” “how to automate onboarding,” “how to sync payroll changes into finance.”

Create a forward deployed motion. The winners will help customers implement. Not as generic consulting, but as a repeatable deployment motion that feeds back into product.

Rethink pricing. Per-seat pricing will not disappear overnight, but it will become less central. If agents consume the product, then pricing needs to reflect usage, actions, workflows, compliance coverage, outcomes, or platform access.

The browser seat becomes an admin seat. The real monetization shifts to governed actions, compliance coverage, and deployment leverage.

Vendors who move fast can own the new layer. Vendors who wait may discover that smaller teams have rebuilt the customer experience on top of their APIs, and captured the relationship.

The pricing question

Nobody knows the final pricing model yet. Anyone who says they do is probably selling something.

But a few models are worth testing:

The best model will depend on the category. ATS might price per hire or workflow volume. Payroll might price by employee, jurisdiction, and payroll run. Customer support might price by resolution. Internal tooling platforms might price by usage and governance tier.

The principle is simple: price where the customer feels the value, not where the old SaaS spreadsheet had a seat count.

The honest risks

There are good reasons to be cautious.

First, most companies are not good at building software. AI makes building faster, but it does not automatically create good architecture. A badly assembled internal OS can become a mess of brittle integrations, undocumented workflows, and invisible dependencies.

Second, agents introduce new risk. If an AI system has authority to act, mistakes can compound quickly. A wrong recommendation is one thing. A wrong payroll action, compliance filing, candidate rejection, or customer refund is another.

Third, vendors may simply recreate lock-in at the API layer. If customers build deeply around one provider’s primitives, they may escape the UI only to become dependent on the backend.

Fourth, the FDE model can become expensive. If every customer requires heavy custom work forever, the company becomes a consulting business with software margins in the pitch deck only.

These are real risks. But they do not invalidate the thesis. They define the execution bar.

The winners will be the companies that combine strong primitives, open standards, excellent documentation, reusable recipes, safe agent controls, and a deployment motion that becomes more productized over time.

The opportunity

The opportunity is not to kill SaaS.

The opportunity is to rebuild where SaaS lives.

For the last twenty years, software companies competed to own the interface. The next generation will compete to own the trusted primitives underneath the customer’s operating layer.

That means the most interesting companies may not look like traditional SaaS companies. They may look like infrastructure companies with domain expertise. Or compliance companies with APIs. Or workflow companies with recipe libraries. Or deployment companies that slowly productize the patterns they see in the field.

Software is losing its head. But the body underneath may become more valuable than ever.

For customers, especially medium-sized businesses, the promise is powerful: keep the hard parts you cannot safely rebuild, and own the layer that makes your business distinct.

For vendors, the warning is just as clear: if your product’s main moat is that users are trained to click through your screens, the clock is running.

The smaller companies that understand this shift will not attack you by copying your UI. They will attack by making your UI irrelevant.

Software is losing its head. But the body underneath may become more valuable than ever.

Subscribe now

We’ve all been there.

You’re scrolling Twitter at 10 AM when OpenAI drops some new reasoning model, and suddenly your timeline becomes a war zone of benchmark screenshots. Then Google fires back with a Gemini update that has a context window longer than a textbook, and Anthropic sneaks in with a Claude version that writes better poetry than your ex.

As expected, the group chat goes wild. I’m usually that one person backing Google—call me a fan if you like, but I honestly think Gemini is very solid, and I stand by it

But here’s the thing: objectively, it’s complicated. ChatGPT still has the cleanest integrations and the widest ecosystem. When you need something to just work, it usually does. Gemini (my personal favourite) is incredibly strong with images, audio, and documents, and if you already live inside Google Workspace like I do, it feels like having a colleague who actually understands your emails and remembers context. And then there’s Claude, which I’ll admit, writes in a way that feels human. Not like a corporate memo generator. It has feeling. Claude’s also great at coding.

But we’re all trapped.

The Golden Handcuffs Are Real

My partner Osita gets it. He sees what Gemini can do. He’s watched me use it. He knows it’s brilliant. But switching? That’s not easy.

He has spent a long time teaching ChatGPT how he thinks, his business context, even the way he prefers sentences to be structured. Moving to another model isn’t just “trying something new.” It’s like starting therapy again with someone who knows nothing about his past. He’d have to explain everything from scratch.

His context is locked up. His digital identity is being held hostage by a company that has zero incentive to let him leave.

And that’s the problem with all of us. We’ve poured so much of ourselves into these LLMs, our writing styles, our preferences, our private thoughts, our work history, that they’ve become more than tools. They’re identity layers. They know us better than some of our friends do.

And we can’t take any of it with us.

Enter the LLM Identity Portability Protocol (LIPP)

This is where I think we need something I’m calling the LLM Identity Portability Protocol or LIPP for short.

Imagine this: a decentralized system where your context lives independently of any single company. All your conversations, preferences, patterns, and personal data exist in a format that you control. Your “AI soul,” for lack of a better term, becomes portable.

On Monday, ChatGPT has the best reasoning for your coding problem? Plug your ‘Soul’ in there. On Wednesday, Claude writes better marketing copy? Slide your context over. Friday, Gemini drops a feature that perfectly fits your workflow? You’re already there, and it already knows you.

No rebuilding. No re-explaining. No starting from zero.

For us, the consumers, this is a total no-brainer:

• Freedom to choose the best tool for the job. You’re not married to one provider because they happen to know your life story. You use whoever’s best at whatever you need right now.

• True data ownership. Your context isn’t some corporate asset sitting in OpenAI’s servers. It’s yours. You decide who gets access, for how long, and for what purpose.

• No more context tax. That psychological barrier, “I’ve invested too much to switch” , disappears. You can experiment freely without feeling like you’re abandoning a relationship.

Why This Doesn’t Exist (Yet)

Of course, the AI labs will never voluntarily build this.

To Sam Altman or Sundar Pichai, your context isn’t just “data”, it’s their moat. If you can leave ChatGPT as easily as you switch from Spotify to Apple Music, they lose their grip on you. They want you locked in. They want you to feel that “Golden Handcuff” tighten every time you think about trying a competitor.

But if these labs actually cared about the user experience, they’d provide Context/Identity APIs. With proper Context APIs or Identity APIs, third-party developers could build personal context vaults that work across models. You grant permission, and any AI instantly understands who you are and how you think.

Here’s the part they rarely say out loud: we’re not just users.

Every prompt we write, every correction we make, every long conversation we have helps train these systems. We’re actively contributing to their intelligence.

In many ways, we’re co-creators.

So it’s strange that the value we help create, our digital identities, our context, our AI “souls”, gets locked away behind proprietary walls.

Your soul shouldn’t be trapped in one company’s file format.

Next time, I’ll talk about the massive opportunities that open up once Identity APIs become real. Because the moment context becomes portable, an entirely new AI economy appears.

But for now, I just want my AI soul 😩.

In 1998, Jamie Dimon was unceremoniously fired from Citigroup. He was the heir apparent, but a power struggle with his mentor, Sandy Weill, left him ousted and publicly humiliated.

He was disappointed. He was also rich. He could have retired.

Instead, two years later, he took the CEO job at Bank One, a troubled regional bank in Chicago. Then he did something truly bold. He took $60 million of his own money—a massive chunk of his net worth—and bought 5.7 million shares of the bank he had just been hired to fix.

This wasn’t just an investment. It was a statement. It was Dimon, fresh off his biggest failure, putting his entire reputation and fortune on the table. He was signaling to every employee, every investor, and every analyst that his fate was now welded to theirs.

He had skin in the game. We know how that story ended: Bank One turned around, merged with JPMorgan, and Dimon became one of the most formidable bankers of his generation.

The Currency of Focus

We tend to think “skin in the game” just means money. It’s the easiest thing to measure. But real skin in the game is often paid in currencies that are far more valuable: time and focus.

When Steve Jobs returned to a dying Apple in 1997, the company was making dozens of beige boxes. They had a sprawling, unfocused product line and were 90 days from bankruptcy.

Jobs didn’t just invest money. He put his entire legacy—the one he had already built at Apple, NeXT, and Pixar—on the line. His first move was an act of brutal, sacrificial focus.

He famously walked to a whiteboard and drew a 2x2 grid: “Consumer” and “Pro” on one axis, “Desktop” and “Portable” on the other. “This is our product line,” he declared, slashing everything else. He said “no” to good ideas so he could say “yes” to four great ones.

That was his skin in the game. He staked his reputation on the belief that focus, not diversification, was the only way to win.

If that’s not enough, consider Marie Curie. She didn’t just invest time in her research on radioactivity; she invested her life. She and her husband, Pierre, worked for years in a drafty shed, painstakingly isolating radium from tons of pitchblende. They suffered from radiation burns and chronic illness.

Today, more than 100 years after her death, her notebooks are still so radioactive they have to be stored in lead-lined boxes. That is existential skin in the game. She paid for her discoveries with her own health, betting her one life on the pursuit of knowledge.

The Dilemma of the Open Door

These stories feel epic, almost foreign. Because today, we are taught the exact opposite lesson.

The modern dilemma, especially for young people, is a world that preaches “structured aimlessness.” We’re told to keep our options open. Build a portfolio of side hustles. Job-hop to climb the ladder. Quiet quit to protect your energy.

The narrative is: Don’t commit.

This is a culture that fears risk. It’s an endless buffet of options, gig economy projects, and half-interests that prevents us from ever having to choose one thing and go all in. We hedge. We diversify our time, our focus, and our identities.

But this “safety” is an illusion. It’s a strategy that protects you from a big, spectacular failure, but it guarantees a thousand tiny ones. It’s a recipe for mediocrity.

You can’t hedge your way to an outlier outcome. You can’t get a 100x return on a 1% investment of your focus.

The only way to make something out of life, to build something that lasts, to truly master a craft, is to put your skin in the game. It demands courage in the face of uncertainty. It requires you to “burn the ships,” leaving no path for retreat.

Boldness rules.

My Fourth Rodeo

This isn’t just theory for me. It’s the only way I’ve ever known how to operate.

I’ve always wanted to build an enduring company. Since leaving college at 19, I’ve built three companies, all with varying levels of failure and success. Each one taught me something. Each one demanded a piece of me.

Now, I’m on my fourth rodeo.

I recently quit my safe, comfortable job to jump back into the ring. To focus. To build again, from the ground up.

Am I certain of a glorious outcome? Not at all. But I am more equipped than ever before. Time will tell the rest.

What I am certain of is that this is the only path. You can’t phone it in. You can’t do it as a side project. It requires your full attention. It requires courage.

Whatever is worth doing is worth doing well. And to get outlier outcomes, you have to take outlier risks. You have to be willing to fail.

You have to put your skin in the game!!.