The daily publication · November 5, 2026 · Small business

Only 1 in 10 big firms let AI decide without a human. The other 9 are the standard.

By Tabron Merryhill · The Workflow

TCS released its Enterprise AI Study on November 4. It surveyed 1,333 senior leaders across 25 countries and 12 industries, at firms with $500 million to more than $20 billion in revenue.

84 percent rank productivity as the leading value from AI. Only 10 percent let AI agents execute decisions without human approval across multiple functions. Only 9 percent say their agents can securely orchestrate workflows across enterprise applications.

Security, risk, and governance outrank cost, speed, and ease of deployment as the factor shaping where AI goes. 39 percent of firms still cannot say whether the value covers token and compute costs. Only 7 percent have an organisation-wide plan to switch providers if one fails, even though 90 percent call model independence important.

TCS's Amit Kapur, who leads the firm's AI and services transformation work, put it plainly. Security concerns and governance gaps are legitimately holding firms back.

The idea

The approval gate is now the measured norm, not the cautious exception. The biggest firms in the world are getting real value from AI and still keeping a human on the decision. That is the standard to copy.

The gap in the numbers tells the story. 84 percent see productivity gains. 10 percent grant autonomy. The distance between those two figures is the approval step. Firms are not waiting to use AI. They are using it with a person in the loop.

That is exactly the shape a small shop should copy. The agent drafts. The agent suggests. The agent prepares. The human decides. Nothing that moves money, changes a price, or makes a promise to a customer goes out without a named person signing it.

The study has a second warning tucked inside it. Only 7 percent of firms have a plan to switch AI providers if one fails, while 90 percent say model independence matters. Owners should hear that too. If your whole workflow runs through one vendor's agent, you do not have a workflow. You have a dependency.

The example

Picture a shop owner who lets an agent handle reorders. The agent watches stock, drafts the purchase order, and flags the items running low. That is good work. It saves hours every week.

Now the agent places the order on its own. It picks the supplier. It commits the cash. It promises a delivery date to a customer. One wrong pick and the owner is explaining a stockout she never approved.

The 90 percent in the TCS study would not run it that way. The agent would draft the order and stop. The owner would check it, sign it, and send it. Same hours saved. Same speed. The decision stays with the person who owns the result.

That is the whole workflow in one line. The agent does the prep. The human does the deciding. Write it down, name the decider, and the approval step stops being a bottleneck and starts being the business.

Here is the honest counterpoint. Komodor surveyed 1,000 engineering leaders the same week and found 84 percent already run fleets of 10 or more operational agents, with more than two thirds spending at least $10,000 a month on them. Agents are already running at scale behind approval rules, not waiting for them. And TCS itself reports 39 percent of firms still cannot say the value covers token and compute costs. So read the 84 percent productivity figure as reported value, not settled profit. The gate is the norm. The payoff is still being measured.

The next step

This week, write the approval rule down for your own shop. List the three decisions an agent may never take alone: money out, price changes, and promises to customers. Put one name next to each one. That name is the approver.

Then check the log once a week. What did the agent draft? What did the approver change? If nothing ever changes, the approver is not reading. If everything changes, the agent is not ready. Either way, you now know, because the decision is on the record.

Want help finding the one bottleneck worth handing to AI first? A free 10-minute workflow conversation will map where the hours go and who signs off before anything final. No pitch, just the map.

Sources: TCS (Enterprise AI Study, primary), https://www.tcs.com/who-we-are/newsroom/press-release/productivity-top-benefit-tcs-enterprise-ai-study; Storyboard18, https://www.storyboard18.com/brand-marketing/only-10-of-enterprises-let-ai-agents-make-decisions-without-human-approval-finds-tcs-study-114008.htm; Express Computer, https://www.expresscomputer.in/news/tcs-study-finds-only-10-of-enterprises-allow-ai-agents-to-act-without-human-approval/140941/; CXOToday, https://cxotoday.com/others/security-and-sovereignty-drive-ai-guardrails-tcs-survey-exposes-enterprise-autonomy-gap/. Opposing view: Komodor, https://www.globenewswire.com/news-release/2026/11/04/3395406/0/en/komodor-survey-finds-more-than-50-of-engineering-organizations-face-control-and-governance-issues-as-they-scale-agentic-operations.html.