The Squeeze in the Middle: How SaaS CEOs Survive the AI Deployment Era

The most dangerous place to be in technology right now is the “messy middle” - That stage of growth where you are neither small and fast enough to behave like the new AI-native startups emerging, nor large and capitalised enough to operate with the force and reach that hyperscalers and foundation model companies are now bringing directly to enterprise customers.

I was recently reading an article from the Mindstudio team on why Anthropic and OpenAI are copying Palantir's Forward-Deployed Engineer playbook, and what struck me wasn't the idea itself (it felt obvious in hindsight) but how familiar it was from my own earlier career. 

We used to call it the "Trusted Advisor" model: if you embed yourself directly alongside the customer, close to their systems and data, you build things that are not only more accurate but far more valuable in practice. Sit at a distance and solve in abstraction, and you almost always miss what actually matters.

That same idea has come back in a much more intense form. If you are trying to deploy AI into large enterprise environments, the difficulty lies in making it work inside organisations full of legacy systems and complicated workflows. The deployment problem is now, in many cases, harder than the model problem itself.

That's where the structural squeeze emerges.

On one side, hyperscalers and AI labs: extremely well capitalised, embedding forward-deployed engineers directly into customers, effectively becoming operational partners who help design and implement entire systems. 

On the other side, AI-native startups: small teams of exceptional people, no legacy, no technical debt, very little organisational friction, able to out-execute much larger companies through focus and speed alone.

Right now, a lot of traditional SaaS companies are caught between those two opposing forces. You cannot match the capital intensity of the hyperscalers. You cannot always match the raw speed of the smallest AI-native teams. Which means the classic SaaS playbook (build software, scale ARR, reduce churn, push self-serve, raise capital, expand upmarket) starts to feel a lot less reliable than it once did.

As AI moves deeper into core enterprise workflows, everything that was previously abstracted away comes back into play: internal systems architecture, data quality, compliance requirements, workflow design, organisational incentives, and how people actually behave inside companies. The gap between what looks impressive in a demo and what works in production is much larger than most people expect.

For SaaS CEOs, this forces an uncomfortable but important rethink across four areas:

The first is talent density, because AI is basically amplifying the difference between exceptional people and average people in a very dramatic way. What that means in practice is that organisations that are overcomplicated, too layered, and too heavy in coordination overhead are going to struggle more and more to keep up. In essence you end up in a world where smaller, flatter, faster teams with very high talent density will consistently outperform larger but slower organisations.

The second is focus, because in moments like this the instinct is often to expand, to become broader platforms, to enter adjacent markets, to hedge across multiple bets. But what tends to win in these environments is the opposite, which is going deeper into fewer, more mission-critical workflows where you genuinely understand the customer better than anyone else and where that understanding becomes a real competitive advantage rather than just a feature set.

The third is services, or more broadly the relationship between software and deployment. For a long time SaaS companies were rewarded for stripping out services in order to protect margins and scale efficiently, but in enterprise AI that separation becomes much less clean. Customers now need help actually making systems work in their environment, which means the real question is not whether you have services or not, but how you design that capability in a way that is lightweight, highly leveraged, and deeply embedded. The use of small forward deployed teams supported by tools, automation and increasingly AI agents that extend their reach rather than just scaling headcount.

And then finally, there is something about recognising what you already have, because most SaaS companies already sit on very real advantages. Whether that’s long-standing customer relationships, deep domain knowledge, embedded workflow understanding, trust, distribution, or proprietary data, you just need to understand what your core advantage is. Ask yourself whether those advantages are actually understood and really appreciated by the customer in a way that makes you indispensable.

The foundational challenge for SaaS CEOs right now is evolving quickly enough not to be outpaced by smaller, more aggressive AI-native teams, while resisting the temptation to copy the hyperscalers, because that is not where traditional SaaS companies will win.

The real opportunity is to redesign how the organisation integrates with the customer: how close it is to the workflows, how fast it learns from real deployment, and how effectively it turns that proximity into compounding advantage. The next generation of great software companies will look quite different from the last, not necessarily because they are building radically different products, but because they are fundamentally more embedded, more talent-dense, and far less attached to the idea that SaaS has to look a certain way in order to scale.

Questions for Reflection:

Talent Density & Execution: How can we radically restructure to maximize talent density and create smaller, faster teams that consistently out-execute slower competitors?

Focus & Indispensability: Are we ruthlessly focused on the few mission-critical customer workflows where our deep understanding creates an indispensable competitive advantage?

Customer Integration & Services: How do we redesign our deployment services to be lightweight, highly leveraged (using AI agents), and deeply embedded to turn proximity into compounding advantage?