Big Business Choking on AI Agents

iantoons.com

Most businesses attempting to integrate AI are discovering the same thing: the technology is rarely the problem. The obstacle is the organization itself — the informal power structures, the processes people have built professional identities around, the quiet resistance to anything that redistributes how work and influence flows. The companies pulling furthest ahead are not transforming. They are building from scratch, and the results are beginning to appear in industries where nobody expected them.

I recently spoke to Toby Simmons, CEO of ActiveXplore Travel, which provides wellness and sport retreats. The operators he works with, retreat hosts and camp leaders, actively encourage their audiences to detox and yet the industry is beginning to be driven by the AI-powered systems they publicly question.

As Simmons observes: "Ironically the wellness industry is about detaching from screens, and yet it is AI that is allowing it to become more accessible."

Built With AI vs. Bolted Onto It

Though Simmons only launched ActiveXplore last October, he is projecting his first million in revenue within months. His former employers with teams of 65 to 70 people are watching sales slide.

The reason isn't hard to locate. Simmons built the business around AI from day one, not as an add-on, but as the operational infrastructure. The system updates pricing automatically when retreat centers change their rates or flight costs shift. Trust accounts self-allocate to bookings. Bespoke client quotes go out within 12 hours and are typically confirmed within 48. As Simmons puts it: "Every company I worked for pushed back on AI, and now they're all chasing their tails trying to work out how to integrate it into a team of 70 people who don't know how to use it - whereas I know full well, when we expand our team, that it's part of the workflow not something employees have to try and learn."

The advantage Simmons has isn't that he's a better technologist than his former employers. It's that he never had to change anything. The obstacle for established organizations attempting the same transition is rarely the technology. It is the unseen internal dynamics. As a CEO, I was guilty of underestimating exactly this, and it's something we confront regularly at HumanDynamics when working with companies on AI transformation. Starting clean removes an entire category of problem. The difference between building AI into a blank workflow and retrofitting it into a functioning organization is not technical. It is human.

Building AI First Companies

Simmons's case is instructive precisely because he started alone. Ciaran Finn and Evan Carroll, founders of performance marketing agency Linear, built with a team — and the numbers tell a different story about what agentic AI does to an existing workforce.

Linear handles performance marketing for e-commerce and software brands: digital advertising across Google, Meta and other platforms, with a heavy emphasis on creative production. The volume and quality of ad creative is, as Finn puts it, "arguably the biggest lever you have to outperform a competitor." Getting that right at scale, quickly, used to require hours of manual analysis per account — identifying which ads were underperforming, briefing the creative team, producing replacements, uploading them. Linear has automated roughly 80 to 90% of that process. Their custom-built system, developed using Claude Code, monitors live ad account data across all clients, identifies creative in need of a refresh, generates replacement assets, and uploads them to Meta. What took hours takes minutes.

The same logic applies across the business. Client sentiment is tracked automatically. Internal reporting runs without anyone manually compiling it. Trend analysis across date ranges, which used to require significant analyst time, is now handled by AI that identifies both underperformance and opportunity faster than human review could. "We're able to spot trends faster," Carroll says, "which means we can turn off something if it's not working much sooner and save a lot of budget from being wasted. Vice versa, if something is working, we can increase the budget on that to get better results for clients sooner."

According to Finn since incorporating AI, Linear’s revenue per team member has gone up 2.5 times and profitability has tripled. Rather than pocket the margin, Linear reinvested it in higher-caliber hires. Average salaries more than doubled. Finn’s explanation: "In the past, if we hired someone on a very high salary who was really skilled, because they're doing everything manually, the amount of effect they can have isn't that high. Whereas now with AI, we can leverage that person's talent much more - they can help more clients, or go a lot deeper with them." The lower-level manual work has been automated making senior expertise more valuable.

Why Agentic AI Is The Difference

Both Simmons and Finn built agentic systems - AI that owns defined slices of work from trigger to completion, without waiting to be prompted. The productivity gains are real but bounded. The structural advantage comes when the agent owns the workflow rather than supports it.

Ayten Hajiyeva, a doctoral colleague of mine at Warwick Business School whose research focuses on human-agentic AI collaboration, has identified precisely why this transition stalls in larger organizations. Her diagnosis, drawn from dozens of in-depth interviews in safety-critical functions across international enterprises, is that agentic AI faces the same problem as a talented new intern: everyone can see the potential in week one, and three months later they are still summarizing meetings and cleaning data. Not because anyone wants to waste them — but because nobody has clearly defined what they are actually allowed to own.

Her framework is practical. Map where AI is already influencing real decisions inside your organization. Pick one decision — repetitive, bounded, low consequence if it goes wrong — and hand it over fully. Redesign the workflow around the agent rather than bolting the agent onto the existing process. Assign explicit human ownership for that decision. Her conclusion: "Real progress only starts once that ownership is explicit. Until then, agents stay stuck at the edges." What Simmons did instinctively when building from scratch has to be engineered deliberately in organizations where workflows already exist and people already have stakes in them. That is the structural disadvantage established companies face.

Most organizations are not behind on AI. They are behind on the decision to trust it — and that decision is considerably harder when you are retrofitting ownership into a workflow that was never designed for it. Smaller companies that build agentic AI in from the start don't have that problem