AI that finishes the work: why the chatbot was only the demo

Eight in ten professionals say AI makes them more productive. Fewer than four in ten companies can find it in their profit. The difference is whether AI helps people do the work, or does the work and lets people review it.

McKinsey’s State of AI survey for 2026, published in August, holds two numbers every leadership team should read together. Eighty percent of respondents say AI has improved their own productivity. Thirty-seven percent say it has contributed anything at all to their organisation’s EBIT, about the same share as a year earlier.

The productivity is real. It just isn’t arriving in the accounts.

Where the saved minutes go

When AI helps someone write a memo faster, the time saved scatters: a shorter afternoon, an extra meeting, a slightly better email. None of it shows up as lower cost or higher revenue, because the work itself hasn’t changed. The same person still assembles the same document through the same steps, only a little quicker.

MIT’s Project NANDA reached a similar verdict in 2025. General-purpose AI tools lifted individual productivity, but just 5% of integrated enterprise pilots were extracting real value; the rest stalled with no measurable effect on profit and loss. The researchers’ explanation wasn’t weak models. It was systems that didn’t retain feedback, adapt to the company’s context or fit how its work actually flows.

The unit that matters is finished work

So change the unit. Stop measuring AI by the tasks it assists and start measuring it by the work it finishes.

An assistant helps a bid manager write a tender response. An engine produces the tender response: complete, in the company’s format, every claim linked to the document it came from. The bid manager reviews it, challenges it and signs it off.

The difference is who assembles. In the first model, people still do the assembling and AI makes them a little faster at it. In the second, the engine does the work and people do what only they can do: judge it.

McKinsey’s high performers, the 6% of respondents who attribute at least 5% of EBIT to AI, point the same way. Nearly three-quarters of them say they have fundamentally redesigned workflows because of AI, up from 55% a year earlier. Among everyone else, the figure is one in four.

What changes for the people involved

In a finished-work model, senior people stop building and start reviewing. The finance director reviews the board pack instead of compiling it from five systems. The head of legal checks the exceptions the engine has flagged in a contract instead of reading every clause against the standard. The marketing lead approves the quarterly plan instead of chasing the inputs for it.

That isn’t a smaller job. It is the job those people were hired for, with the assembly taken off their desks.

How to choose the first piece of work

Most companies start with the most exciting use case. The better start is the most repetitive one. Four tests help pick it.

  1. It recurs. Monthly, quarterly, or with every new client or tender.
  2. It runs on documents the company already has. Policies, past proposals, financials, contracts, price lists.
  3. It follows rules someone could write down. A format, a checklist, a standard to compare against.
  4. It has a clear reviewer. Someone senior already signs it off today.

Quarterly plans, tender responses, contract checks, board reports, audit preparation and sales or marketing strategies usually pass all four. Score each candidate from 0 to 2 on each test, then start with the highest score and the most frustrated reviewer.

Mid-sized companies have an edge here. MIT found mid-market organisations moved from pilot to full implementation in about 90 days, against nine months or more for large enterprises.

Why finished work depends on the other two decisions

An engine that finishes work needs two things underneath it. It needs governed data, or it will finish the wrong work with total confidence. And it needs a private home in the region, or the legal team will stop the project before it starts. That is why the first two pieces in this series dealt with data readiness and where AI runs: they are the ground finished work stands on.

What this looks like by 2028

Our forecast: within two years, job descriptions across the region will start to say “reviews” where they now say “prepares”. Companies that make that shift deliberately, one recurring piece of work at a time, will grow output faster than they grow headcount. Companies that keep buying assistants will keep finding the productivity in surveys and not in their accounts.

That is the work Unknot Technology Labs Ltd builds for. Unknot Engine takes on the work that slows a company down, from quarterly plans and tenders to contracts, board reports and audits, and delivers it complete, with every source shown, ready for approval. Your people review it. The engine does the rest.

Ahmed A. H. Hamid is the founder of Unknot Technology Labs Ltd at the DIFC Innovation Hub in Dubai. He spent more than eight years in business development and strategy consulting with medium and large enterprises and professional-services firms in the UAE and Saudi Arabia.

Sources: McKinsey, The state of AI in 2026: On the road to ROI (August 2026); Virtualization Review on MIT Project NANDA, The GenAI Divide (August 2025).

See it on ten of your own documents.

A 45-minute session at your office. Bring contracts, policies or price lists; leave with answers, finished work and their sources.

We reply within one business day.