What does ICONIQ's State of AI 2026 report mean for enterprise AI adoption?

ICONIQ's 2026 report surveyed AI builders, not enterprises adopting AI, so read it as a preview, not a benchmark. Software companies that build AI products are redesigning roles, still need people to step in on many agent tasks, and report modest agent productivity gains. The lesson for enterprise leaders: value depends on who owns the work, not the tool.

Start with who answered the survey, because it changes how you should use every number in it. ICONIQ's State of AI 2026 report asked about 300 executives at software companies that build AI products. Most were in North America. It is not a survey of UK enterprises, and it is not a global view of AI adoption.

That still makes it useful. These companies build AI for a living and use it inside their own organisations. What they report about roles, oversight and speed is an early signal of what enterprises adopting AI may face next.

Who did ICONIQ survey, and why does it matter?

ICONIQ surveyed about 300 executives in Q2 2026, mainly CEOs and heads of engineering, AI and product, at software companies building AI products. It compared the results with a similar survey of about 300 executives in Q4 2025 (ICONIQ, 2026).

Respondents were 87% North America and 13% Europe. The sample also leans towards technical and product leaders rather than HR, finance or operations.

So treat the findings as a view from the front of the queue. If the companies that build AI for a living are still working out who owns what, an enterprise that is earlier in its AI programme should expect the same questions, and should plan for them now.

What did the report find about how AI builders organise work?

The headline is that these companies are changing the shape of their organisations, not just adding tools. AI has led 78% of the companies surveyed to rethink how they plan their workforce (ICONIQ, 2026). 45% expect a different mix of roles, and 33% expect smaller teams.

Structure is shifting too. Where AI brings in most of a company's revenue, 72% keep four or fewer layers of management from the CEO down to the most junior individual contributor (ICONIQ, 2026). Elsewhere in the sample, 56% do.

Speed of adoption separates the leaders from the rest. The top performers take roughly 2.5 months to get a new AI tool live (ICONIQ, 2026). Their peers take about 3.5 months.

What did it find about AI agents and human oversight?

Agents are working, but they are not working alone. Close to half of the companies surveyed report that a person still has to intervene in 30% or more of their agents' tasks (ICONIQ, 2026).

Agents fail most often on multi-step workflows that break partway through. That is exactly the point where someone has to notice, decide what happens next and finish the work.

Productivity gains are real but modest. In every revenue band in the survey, the average productivity gain from agents is under 30% (ICONIQ, 2026). For a board expecting step changes, that is a useful correction: agents add capacity, and people still carry the outcome.

Is the gap technology or leadership?

In our view, it points to leadership. A separate Gallup study of US employees found that 47% say their organisation has begun integrating new AI technology or tools (Gallup, 2026). Yet only 25% strongly agree that their employer has set out a clear AI plan or strategy.

Read together, the two studies point the same way. The AI builders in ICONIQ's sample are redesigning roles and still need people to step in on agent work. Meanwhile, most US workers do not strongly agree that their organisation has communicated a clear AI plan.

In our view, that is not a technology problem. It is a question of who decides, who owns the result and who explains the change to the people doing the work.

What is Generation Digital's view?

Our view: buying AI is the easy part. In our view, the gap between AI spend and AI value is a people gap. Value shows up when a named person owns each AI-supported outcome, when teams agree in advance which steps a person checks, and when leaders tell people plainly what AI is for and what it is not for. Without those three, tools sit on top of old habits.

This is a statement of how Generation Digital approaches adoption, not a finding from the ICONIQ or Gallup data. It is why our AI adoption services start with owners and use cases before features and agents.

What should enterprise leaders do next?

Use the report as a checklist of decisions, not a set of targets. Each step below has a named owner, because a step without an owner tends not to happen.

If you already run Claude across teams, an optimisation review is a quick way to check whether ownership and oversight have kept pace with usage. For larger rollouts, see how we support Claude Enterprise.

What should you avoid when reading this report?

Do not present the numbers as a UK benchmark. The sample is mostly North American AI builders, so quoting it as 'what enterprises are doing' will mislead a board.

Do not read modest agent gains as a reason to wait. The point is that agents need a person directing them, so design for that from the start.

Do not treat a smaller team as the goal. A different mix of roles was the more common expectation in the survey, and, in our view, the harder change to lead well.

Book a Claude briefing

If you are an exec sponsor deciding who owns AI in your organisation, book a Claude briefing with Generation Digital. We will help you set owners, guardrails and a plan your people can act on.

FAQ

Is the ICONIQ State of AI 2026 report a survey of enterprises adopting AI?

No. ICONIQ surveyed about 300 executives at software companies that build AI products, mostly CEOs and heads of engineering, AI and product. 87% were in North America and 13% in Europe. It shows how companies that build AI products run AI internally. Enterprise leaders should read it as an early signal of what is coming, not as a benchmark for their own sector or country.

What does the ICONIQ report say about jobs and team structure?

In ICONIQ's 2026 survey, AI has led 78% of companies to rethink workforce planning. 45% expect a different mix of roles and 33% expect smaller teams. Where AI brings in most of the revenue, 72% of companies keep four or fewer management layers, against 56% of the rest. The message for a CHRO is that role design, not headcount alone, is the main change to plan for.

Do AI agents still need human oversight, according to the report?

Yes. Close to half of the companies in ICONIQ's 2026 survey report that a person still has to intervene in 30% or more of agent tasks. Multi-step workflows that break partway through are the most common failure. In every revenue band, the average productivity gain from agents is under 30%. Agents add capacity, but a named person still needs to own each outcome.

What stops AI rollouts delivering value?

In Generation Digital's view, it is often because nobody has explained the plan or named an owner. Gallup's 2026 data shows 47% of US workers say their organisation has begun integrating AI tools, but only 25% strongly agree a clear AI plan has been communicated. Value comes when a named person owns each outcome and leaders say plainly what AI is for.

What should a CEO do first after reading this report?

Publish a short, plain AI plan and name an owner for each AI-supported outcome. Then ask each use-case owner to decide which steps a person checks, and ask the CHRO to review how roles and team shape will change. These are leadership decisions, not technology decisions, and they apply whichever AI platform your organisation uses.

Last updated: 05 Nov 2026

By Graham Mackay, CEO, Generation Digital