Every enterprise will eventually be augmented by AI agents. That part isn’t controversial anymore. The question nobody is answering well: what makes this transformation succeed?
The Generic Trap
Ask any AI team what they’re building and you’ll hear the same list: chatbot for customer service, document search, report generation, internal knowledge base. These use cases are real — and completely generic. A bank, a railway company, and a hospital will describe nearly identical projects.
This should be a warning sign. If your AI strategy looks the same as every other company’s, you don’t have a strategy. You have a technology shopping list.
The Economics Aren’t There Yet (For Most)
Here’s what most enterprise AI leaders won’t say publicly: the ROI on GenAI agents is unproven for the majority of use cases. Inference costs are high. Reliability is low. The gap between a compelling demo and a production system that handles edge cases is enormous.
Traditional ML has clear wins — fraud detection, credit risk, demand forecasting. These models have been in production for years with measurable returns. GenAI doesn’t have that track record yet.
So should enterprises wait? For many, yes. The rational move is to let the technology mature, costs drop, and patterns stabilise. Running expensive pilots that won’t scale isn’t innovation — it’s theatre.
When It Does Work: The Hard Part Isn’t Technical
For the enterprises where the economics do clear the bar, the real challenge emerges: agents need authority to act, not just ability to think.
Most companies bolt AI onto existing processes. The chatbot answers questions but can’t resolve anything. The document assistant summarises but no one changes how decisions get made. The agent drafts reports that humans rewrite entirely.
The transformation that matters is organisational:
- Workflow redesign. Which human steps can an agent own end-to-end, not just assist with?
- Incentive alignment. If agents make a team 3x more productive, does the team shrink or take on harder problems?
- Authority architecture. What can an agent decide autonomously? What requires human approval? Where’s the boundary?
- Institutional knowledge capture. The most valuable enterprise knowledge lives in the heads of long-tenured employees approaching retirement. Agents can preserve it — but only if someone builds the capture process before those people leave.
The Early Adopter Bet
Adopting agents now is a risk-reward calculation. The risk: you spend heavily on immature technology with uncertain returns. The reward: if you get the organisational transformation right early, you compound the advantage as the technology matures and costs drop.
The companies that will win aren’t the ones deploying the most agents. They’re the ones redesigning their organisations to work with agents — changing authority structures, decision flows, and team compositions. The technology is the easy part.
The hard question for any AI leader: are you transforming your organisation, or just buying software?