A business can use AI to write an email without changing how it handles an order, resolves a customer problem or pays a supplier. For Circeus, that gap between access and implementation is where the next phase of business AI needs to focus.
In a new article, “AI is moving fast. Why has adoption been slow in the everyday economy?”, Andrea Mostosi, Head of AI at Circeus, argues that the challenge is not simply building better models or persuading business owners to try them. It is bringing AI into the technology through which their work already happens.
As TFN previously reported, the London-based group has raised more than $220 million in total capital. Its mission is to diffuse AI into the real economy by acquiring mission-critical software businesses and evolving their products into AI-native systems of action.
Across its portfolio, Circeus’s software serves more than 250,000 businesses. Mostosi’s piece sets out why the group believes those existing products and customer relationships offer a route to bring AI into everyday operations.
Using AI is not the same as changing the work
Official figures cited in Mostosi’s article put AI use at between 17% and 20% of US businesses and just under 20% of EU enterprises. The US Census Bureau data also show a pronounced size gap: 37% of firms with at least 250 employees use AI, compared with fewer than 20% of those with four or fewer.
Depth matters too. Among UK businesses with at least 10 employees using AI, around one in ten described that use as extensive, according to the ONS figures cited in the piece.
Circeus’s own assessment is that meaningful workflow implementation in the everyday economy is closer to 4–5%, roughly one in twenty. Mostosi draws a distinction between AI that changes how a workflow runs and AI that simply helps someone write an email.
Better models are only part of the answer
Mostosi does not dismiss model progress. He points to METR’s work measuring the length of tasks AI can complete at a 50% success rate, which has moved from minutes of expert human work to hours on software and technical benchmarks. But, as METR itself cautions, success on a benchmark is not the same as being able to delegate work safely.
The practical obstacle is what surrounds the model. Mostosi’s example is a 40-person distributor without a data team or integration budget. A standalone chat window still leaves someone finding records, pasting them in, checking the result and moving it back into the business’s systems.
“For the owner, that is a new job, not a saved one,” Mostosi writes.
The route to adoption is already there
His argument is that useful AI should arrive inside existing technology rather than as another integration project. Accounting packages, booking systems and order-management software already hold much of the data, permissions and workflow logic needed to make it useful.
That does not eliminate the engineering work. But it provides a foundation to build on, rather than asking every customer to assemble one independently.
The distinction matters for the companies building AI products. Existing software has already been bought, connected, and trusted, often for years. Bringing AI into those products gives providers a starting point that a standalone tool still has to establish.
From AI features to dependable workflows
Circeus describes its approach through three layers: the software knows, the model does and the person decides.
The software supplies the records and access controls. The model works within a bounded set of actions. A human approves when the consequences warrant it. A support agent might prepare a reply, for example, without being authorised to issue a refund independently.
Evaluation sits beneath all three. The group’s stated standard is to track task completion, error rates, human intervention and the cost to each business, then keep retesting as models and workflows change.
“If we cannot measure whether it is doing the job, it does not ship,” Mostosi writes.
Why build across a portfolio?
Circeus’s answer is also an ownership strategy. It acquires software businesses and gives them access to shared AI engineering, rather than expecting every specialist product team to build the same capabilities alone.
A 20-person software business may know its customers’ industry in detail but struggle to maintain evaluation tools, safe data access, guardrails and human approval processes alongside its product roadmap.
The group’s model is to build reusable foundations on a centralised AI platform and deploy them across the portfolio. A support capability developed in one product can then be adapted to another’s customers, data and workflow.
Each portfolio company continues to build products for its own market, while sharing infrastructure and lessons across the group. The aim is to make each new deployment faster and cheaper, without forcing the same product on every business.
Long-term ownership, not a one-off AI upgrade
Mostosi ties this approach to a broader commitment: giving the businesses that underpin the everyday economy a fair chance to compete as AI changes their markets.
That requires working with the people who understand the business, including the unwritten rules, exceptions and relationships that a general model cannot supply on its own. Circeus argues that owning software businesses for decades creates the conditions to keep improving those systems alongside their operators.
“We want the founders who sell to us to become more ambitious about their products, not less, and to have the capital and engineering talent to act on that ambition,” Mostosi writes.
That is the connection between the group’s acquisition model and its AI mission. The aim is not just to put an assistant beside existing software, but to evolve systems that record what happened into systems that carry out more of the work, with people guiding and approving.
“The everyday economy should not have to work out AI on its own,” Mostosi writes.
For Circeus, the goal is to make AI a dependable part of the technology businesses already use, rather than another technology project they have to take on.
Read the full article, “AI is moving fast. Why has adoption been slow in the everyday economy?”, on the Circeus blog.
