Few topics in the SaaS world spark stronger opinions than Product-Led Growth versus Sales-Led go-to-market. And now that SaaS has been battling for survival in public markets, the debate seems to be looking for a new arena. AI adoption is shaping up to be the next battleground.

Top-down versus bottom-up adoption. Should AI transformation be driven by leadership, process redesign and company strategy? Or does real change emerge from thousands of employees experimenting and finding value themselves? And perhaps more importantly: could individuals-led AI adoption lead to company wide transformation?

Kyle Norton, the CRO of Owner.com, wrote a particularly thought-provoking take on this. Looking at implementing AI in his own GTM org, as well as examples from other top GTM orgs, he argues that centralised AI deployment is the only way to get “production-grade results, speed, and the kind of quality and volume that actually moves revenue.”

Speaking with founders of AI Enablement services businesses, we see both approaches deployed in the field. Each has its own advantages and drawbacks and requires making GTM choices and setting the company on a very specific trajectory that’s hard to pivot from. 

Which led me to an interesting question - if I was a founder building an AI Enablement services business today, which approach would I bet my company on?

ChatGPT crossed 100 million users in just two months, becoming the fastest-growing application in history. Lovable and Replit enabled non-technical founders to build and ship products, going from zero to $100m in revenue in less than a year. The most widely shared AI content is rarely about enterprise transformation roadmaps or operating model redesign. It is "10 useful prompts for X", "how to use Claude better", and "AI tools that save you three hours a day".

We saw this firsthand when mapping the AI media landscape while scouting potential acquisitions and investments at 10xHumans. The largest publication we found focused on AI within a specific business function (an "AI in Product" newsletter) was roughly one-tenth the size of the largest productivity-focused publications. In the latter category, reaching 50,000 subscribers was relatively common and no longer a meaningful differentiator. The market for "how AI helps you" appeared to be an order of magnitude larger than the market for "how AI transforms your department."

That should not be surprising. Bottom-up adoption naturally spreads faster because the value proposition is immediate and personal. An employee does not need leadership approval or a change management programme to save thirty minutes writing emails or summarising meetings. 

From the perspective of the organisation, putting AI directly into the hands of the people who actually experience the daily workflow problems can surface highly creative and interesting ideas from the "edges" of the organization. It also encourages company-wide AI literacy. 

But the impact has yet to show up in companies’ P&Ls.

Individual employees rarely build production-grade systems. Most experiments stop at the level of chats, prompts or custom GPTs. These workflows can be highly useful for the individual, but they are often difficult to scale, maintain or integrate into broader company systems.

Another issue is fragmentation. Decentralised experimentation can lead to a Frankenstein collection of artifacts scattered across an organisation. Valuable context gets trapped inside personal systems. A sales representative might build an AI workflow that surfaces important customer insights, but if those learnings remain locked in a personal tool rather than feeding into shared databases or customer success workflows, the broader organisation never captures the value.

But the most important reason is focus. When large numbers of non-technical employees begin spending significant time building workflows, coding or managing AI agents, the day job gets neglected. While experimentation can create useful skills, there is a point where employees risk moving away from the work they were originally hired to do. The salesperson becomes part prompt engineer, the marketer becomes part workflow developer, and the organisation may end up pulling people away from their core competencies rather than amplifying them.

Security and compliance remain the elephant in the room. Even if employee adoption continues to accelerate, there comes a point where IT steps in, concerned about confidential data finding its way into unauthorized cloud platforms.Training providers might think they are immune, but if the very tools they are teaching employees to use get banned, demand for their services can evaporate overnight.

The biggest advantage of centralised AI Adoption is quality and leverage. A dedicated AI team can build systems that are often an order of magnitude more capable than what even highly AI-savvy employees can create on their own. Centralised teams can build around structured data pipelines, retrieval systems, context engineering and evaluation frameworks, keeping compliance and security in mind. 

Reducing tool fatigue is another real benefit. One of the hidden costs of decentralised AI is that every employee effectively has to become a part-time AI operator: learning interfaces, managing prompts and figuring out workflows. Top-down approaches can remove that burden entirely. Rather than asking teams to adapt to new AI tools, the AI itself adapts to existing workflows. Outputs appear directly in the systems employees already use every day: CRM platforms, Slack channels, internal dashboards and other familiar environments. The user does not need to change behaviour because the experience becomes largely invisible, making change management easier.

There is also a deeper opportunity around how work itself gets structured. Rather than simply helping people execute the same jobs more efficiently, centralised AI creates the possibility of unbundling jobs entirely. Brett Taylor, CEO of Sierra and Chairman of Open AI,  spoke about it in his Cheeky Pint interview. Many roles today contain a mixture of high-value human work and repetitive administrative overhead. Centralised AI systems remove much of that operational burden, allowing employees to spend more of their time on the parts of the job that actually require human judgment, creativity or trust.

Building production-grade AI systems requires specialised expertise that sits somewhere between engineering, data science and product design - often described as “applied AI talent”. These people are among the most sought-after profiles in the market. Attracting them requires strong employer branding, meaningful compensation packages and the resources to compete with top tier tech companies. 

Orchestration is another challenge. AI systems become exponentially more complicated as they spread across different parts of an organisation. The difficult part is not building individual systems. It is ensuring they all share the same context and operate from a common understanding of the customer. Managing a fleet of AI agents so they operate from a common understanding of the customer remains a big technical challenge, even when deployment is supported by competent AI consultants.

Building foundational AI capabilities is a lengthy process (as is selling a service to help a customer build these capabilities). Developing core intelligence layers, integrations and infrastructure can take months or years, while consuming substantial engineering resources along the way. While “AI” is top of the agenda for most business leaders, there is a risk that building AI becomes a distraction from building the company's actual product or serving customers. 

If I were building in this space today, I don’t think the answer starts with product or strategy. It starts with founder DNA and the type of motion you are naturally suited to execute.

Are you comfortable operating in a boardroom with senior decision makers? Do you, or someone very close to you, have experience navigating multi-stakeholder enterprise deals? Can you balance the tension between closing a deal quickly and waiting long enough to expand scope and deal size meaningfully? If the answer is yes, you are naturally suited toward a top-down, enterprise motion.

But that path comes with its own requirements. You will likely need a more senior initial team, credible case studies earlier than most startups, and a clear strategy for overcoming trust and credibility barriers. We covered this topic previously.

On the other side, if you are building a consumer or small-team focused AI Enablement service, the game shifts entirely. Success depends far less on account management and far more on marketing instinct: optimising acquisition channels, finding creative organic loops, and building something that spreads without needing formal procurement. 

In both worlds, services are inherently harder to defend, which means distribution and GTM execution become the primary source of advantage.

While most visible AI success stories today are bottom-up in nature, it is still difficult to build a truly large business without large, repeatable customers. Translating bottom-up adoption into organisation-wide deployment is not automatic. It is a different skill that not even many successful product-led SaaS companies have fully solved.

Which is why the most interesting models are often hybrids.

One approach that sits directly between these two worlds is what I would call AI Champion Enablement. Instead of choosing between bottom-up and top-down, it starts at the top but then quickly spreads down. A vendor works with senior executives to identify where AI can create the most leverage across the organisation. From there, internal “AI Champions” are identified and trained to become power users and internal evangelists. These champions then propagate AI usage across the rest of the company, effectively turning executive intent into grassroots adoption.

It is a hybrid motion: top-down in strategy, bottom-up in execution. And in many ways, it may be one of the most effective ways to bridge the gap between individual productivity gains and real organisational transformation.

- Daria