The Second Wave of AI Adoption Looks Nothing Like the First

The first wave of AI adoption bought tools and hoped people would find uses for them. The second looks entirely different and works considerably better. This piece covers the distinction, drawing on FISTA Solutions' AI enablement delivery work.

What separates the two waves?

Almost every structural choice.

First wave	Second wave
Buy a tool, distribute widely	Target one workflow
Adoption as the metric	Process outcome as the metric
Innovation team owns it	Operations owns it
Measure after deployment	Baseline before deployment
Scope expands with enthusiasm	Scope narrows deliberately
Training is a launch email	Change management is funded

Why did distributing tools not work?

Because capability does not organise itself into value.

Giving a capable assistant to ten thousand people produces ten thousand shallow uses and very little measurable change to any process. Individuals save minutes; the organisation's throughput, quality, and cost structure are unchanged.

That is not a criticism of the tools, which are genuinely useful. It is an observation that organisational change requires someone to change something, and buying software does not assign that responsibility. See why ai projects will look like ERP projects.

What does workflow-led look like?

Pick one process, understand it, measure it, change it, and verify.

That means mapping how the work actually happens rather than how it is documented, identifying where the model genuinely helps, integrating with the systems of record, and measuring the same metrics before and after.

It is unglamorous and it produces results that survive a budget review, which the first approach did not. See AI pilot to production.

Why does ownership move?

Because the people accountable for a process are the ones who can change it.

An innovation function can run a pilot; it cannot alter how a claims team works, retrain them, or accept the risk of the new process. Those are operational authorities.

When the operational owner drives the project, the change sticks because it is their change. When a central team drives it, the process reverts once attention moves elsewhere, which is the most common failure of first-wave programmes.

Why is the baseline so important?

Because without it no return can be demonstrated, ever.

Measuring handling time, error rate, or throughput after deployment tells you the current state and nothing about the change. The comparison requires numbers taken beforehand, and once the process has changed those numbers cannot be recovered.

This single omission is why many working systems cannot defend their budgets. The system helped, and nobody can prove it. See how to calculate AI ROI.

What does change management involve?

Working with people whose job is changing, most of whom did not ask for it.

The first wave treated training as a launch announcement. The second treats it as a workstream: explaining what changes, addressing the reasonable concern about job security honestly, retraining for the new work, and adjusting targets that assumed the old process.

Systems fail at this step more often than at any technical one. A correctly working system that people route around has failed. See AI adoption strategy.

What does the second wave produce?

Fewer systems that persist, rather than many that fade.

A workflow-led programme might deploy three or four systems a year rather than distributing one tool to everyone. Each has a named owner, a measured effect, and operational funding.

That is a less impressive slide and a better outcome. It also compounds: each deployment builds integration, evaluation, and operational capability that makes the next one faster.

What is the counter-argument?

The counter is that broad tool distribution produced genuine individual productivity gains that are real but hard to measure, and cancelling it would remove value. That is fair — the two are not mutually exclusive. General tools for general knowledge work, workflow programmes for process change, funded and measured differently.

What does this change for engineering teams?

It changes what engineering teams are asked for: integration with systems of record, evaluation against domain criteria, and operational reliability, rather than prototypes.

It also means engineers work closer to the process owners, which is a different working pattern from building a platform and handing it over.

What does this change for buyers?

It means evaluating vendors on whether they can change a specific process, with evidence from a comparable deployment, rather than on capability.

It also means budgeting for your own people's time, because a workflow change is not something a vendor does to you.

What should leaders do about it now?

Pick one workflow, measure it before anything changes, assign an operational owner, and fund the change management. That sequence is the whole method.

Resist the pressure to demonstrate breadth. Three processes genuinely changed is worth more than thirty pilots.

Where do agents fit in this?

They are the natural second-wave technology, because they act within a specific process rather than assisting generally.

That also makes them harder: an agent needs the workflow understood, the permissions defined, and the exception path staffed. The organisations succeeding with agents are those that did the workflow work first. See AI pilot checklist.

How will you know if this is happening?

Watch for AI ownership moving into business units, for baselines being required before projects start, and for fewer, larger initiatives replacing many small ones. Those mark the transition.

How FISTA Solutions reads this

FISTA Solutions builds and operates production AI systems through AI agents, AI enablement, and forward deployed engineering: one workflow targeted at a time with a baseline measured before anything changes, and an operational owner accountable for the process outcome, decisions documented with their reasoning, and handover that leaves your team able to maintain what was delivered. The record is 150+ projects for 50+ companies across 12+ countries.

To discuss what this means for your roadmap, message FISTA on WhatsApp, or read AI pilot to production.