Most enterprises are getting faster with AI because speed is easy to buy, while transformation needs people to decide who owns what. Deloitte's 2026 survey finds 81% report faster decisions and execution, yet only 25% are deeply transforming. When AI acts on its own, accountability belongs to a named person, and that person should be in place before autonomy widens.
Speed is now the common result of AI adoption. Transformation is not. The gap matters because AI is starting to act without asking, and many organisations have not yet decided who answers for the result.
This article sets out what Deloitte found, why faster work does not add up to a changed business, and the decisions an executive sponsor should make first.
What did Deloitte find?
Deloitte's State of AI in the Enterprise report (2026) is based on a global survey of 3,500 senior leaders. Its headline numbers describe organisations that have moved fast on tools but slowly on the work around them.
Area | What leaders reported (Deloitte, 2026) |
|---|---|
Speed | 81% say AI has improved the speed of decision-making and execution |
Transformation | Only 25% are deeply transforming the business |
Job redesign | 60% have redesigned jobs around AI to at least a moderate degree; only 15% call the changes extensive or complete |
Autonomy | 40% let AI execute decisions or actions without human approval to at least a moderate degree |
Accountability | For 23%, it is unclear who is accountable for a negative outcome |
Cost | 71% track total AI inference and API spend; only 44% track per-unit inference cost |
Revenue | 28% report revenue gains from AI, up from 20% in Deloitte's January survey |
Deloitte's recommendations point the same way. In our words, it asks leaders to redesign the work itself: who owns it, who has authority and where human judgement matters. It also asks them to set approval, monitoring and escalation rules according to what each AI system is allowed to recommend, execute or decide.
Why does speed not add up to transformation?
Speed comes with the tool. A team that drafts, summarises or analyses faster gets that benefit in week one, with no change to who does what.
Transformation needs decisions that only people can make. Which roles change? What does a manager now review instead of produce? Who signs off when an AI system takes an action? The Deloitte survey (2026) shows the gap clearly: 60% have started redesigning jobs, but only 15% have gone far.
The cost numbers tell a similar story. Most organisations can see the total AI bill (71%), but fewer than half (44%) can see the cost per unit of work. Without that view, a finance director cannot tell which uses pay back and which do not.
Revenue is moving, with 28% now reporting gains against 20% in January. That is progress. It also means most organisations still report speed, not new value, as their main return.
Who is accountable when AI acts on its own?
This is the question the report raises most sharply. In the Deloitte survey (2026), 40% let AI execute decisions or actions without human approval to at least a moderate degree. For 23%, accountability for a bad outcome is unclear.
Put those two numbers side by side. Autonomy is growing faster than ownership. That is not an argument against AI agents: agents that act are where much of the value will come from. It is an argument for deciding who manages them.
Where does Generation Digital stand on AI accountability?
Our view: speed is the easy part of AI adoption and transformation is a people decision. Before any AI system is allowed to act without approval, a named person should own the outcome, the decision rights should be written down, and the affected roles should be redesigned so people know what they now decide and what they review. Widen autonomy only when that owner is in place.
This is a point of view, not a survey finding. It follows the direction of Deloitte's recommendations and turns them into a sequence a leadership team can act on. Human-led does not mean slower. It means someone owns the result.
What does this mean for you?
Treat autonomy as something you grant in steps, each with an owner. The table below is a starting sequence for a leadership team.
Step | What to decide | Owner |
|---|---|---|
1. Map AI actions | List every place an AI system recommends, executes or decides today | CIO |
2. Name an owner per decision | One accountable person for each AI-driven decision, written down | Executive sponsor |
3. Set approval rules | What needs sign-off, what is monitored and when to escalate, by level of autonomy | Risk or governance lead |
4. Redesign the affected roles | What each role now decides, what it reviews and what it stops doing | CHRO with line managers |
5. Measure unit cost and value | Cost per unit of work and the commercial result it supports | Finance director |
6. Widen autonomy | Extend only where steps 2 to 5 are complete | Executive sponsor |
For Claude deployments, the rules in step 3 belong in your Claude governance model, set by a workspace admin and reviewed by the governance lead. Step 4 is where AI adoption support and Claude training matter most, because people need to practise the new way of working, not only read about it. Step 5 is an AI optimisation question: knowing which uses earn their cost.
Teams rolling out Claude Enterprise can apply the same sequence from day one, before agents are connected to live systems.
What should you avoid?
Counting speed as transformation. Faster drafting is a real gain. It is not a new operating model, and reporting it as one hides the work still to do.
Granting autonomy by default. If a system can act, someone should have decided that it may. Settings switched on by default are not a decision.
Shared ownership. 'The AI team owns it' means no one does. Name a person, not a function.
Tracking only the total bill. A single spend figure cannot tell you which use cases to grow and which to stop.
Redesigning workflows but not roles. Deloitte separates the two for a reason. A new process with unchanged job definitions leaves people unsure what they now decide.
Talk to us
If your organisation is widening what AI can do on its own, start with who owns each decision. Book a Claude briefing with Generation Digital to work through the sequence for your teams.
FAQ
Why are enterprises getting faster with AI but not transforming?
Speed arrives with the tool, while transformation needs leaders to change roles, ownership and decision rights. Deloitte's State of AI in the Enterprise report (2026) found 81% of senior leaders report faster decision-making and execution, but only 25% are deeply transforming the business. The difference is the people work: deciding who does what once AI is part of the job.
Who is accountable when AI makes a decision on its own?
A named person should be. In Deloitte's 2026 survey, 40% let AI execute decisions or actions without human approval to at least a moderate degree, and 23% said accountability for a negative outcome is unclear. Generation Digital's view is that every AI-driven decision needs one written-down owner before the system is allowed to act without approval.
How much have organisations redesigned jobs around AI?
Most have started, few have finished. According to Deloitte (2026), 60% have redesigned jobs around AI to at least a moderate degree, but only 15% describe those changes as extensive or complete. Redesign means deciding what each role now decides, what it reviews and what it stops doing, not only changing the workflow.
Should organisations limit how much AI agents can do?
They should grant autonomy in steps rather than by default. Deloitte recommends setting approval, monitoring and escalation rules based on what each AI system can recommend, execute or decide. Generation Digital suggests naming an owner for each decision first, then widening what the agent may do once that owner, the approval rules and the redesigned roles are in place. This makes agents more useful, not less.
How should leaders measure the return on AI?
Track cost per unit of work and the commercial result it supports, not only the total bill. Deloitte's 2026 report found 71% track total AI inference and API spend, but only 44% track per-unit inference cost. It also found 28% report revenue gains from AI, up from 20% in its January survey, so value is moving beyond speed.
Last updated: 10 Nov 2026
By Graham Mackay, CEO, Generation Digital
Put this into practice
Rolling out Claude or AI across your organisation?
Generation Digital is a certified partner in Anthropic’s Claude Partner Network. Talk to one of our Claude specialists about rollout, training and adoption.
Get weekly AI news and advice delivered to your inbox
By subscribing you consent to Generation Digital storing and processing your details in line with our privacy policy. You can read the full policy at gend.co/privacy.
