PART III OF III - THE DATA & THE MACHINE
The pitch is compelling. The founders are credible. The investor roster is as strong as we've seen at seed stage. This is the final piece and where the real questions live.
In Part I, we introduced the product - the idea that you can type a sentence and have an AI agent run with it.
In Part II, we told you about the founders who built it and the research they left behind to do so.
Now, in this final part, we put it all together.
This is where we answer the question we've been building toward across this entire series: does Codewords actually work, and if so, what stands between it and the category it's trying to own?
Let's start with what Codewords actually is, mechanically - because as we hinted in Part II, the pivot from research to product wasn't just a change of direction. It produced something architecturally distinct from everything else in the market.
Most no-code automation platforms - Zapier being the most obvious example - work through a visual drag-and-drop interface.
You connect App A to App B, define a trigger, define an action, and the workflow runs. It's powerful for simple, linear tasks.
But change one step, add a layer of complexity, or ask it to handle a genuine business decision with multiple possible outcomes, and the whole chain becomes brittle. It breaks in ways that are hard to predict and harder to debug.
Codewords takes a fundamentally different approach. When you describe a task in plain English - say, 'monitor my competitor's pricing page every three days and alert me when prices change' - the platform doesn't build a visual flow.
It generates Python-based logic.
Actual, readable code, created from your description. Python is more stable for complex, branching workflows. It handles conditional logic - if this, then that, but if something else, then this other thing - without falling apart.
The AI agent that runs this logic is called Cody. Once deployed, Cody runs persistently - on a schedule, triggered by events, or continuously in the background. It doesn't wait to be started each time.
It doesn't need a human to kick it off on Monday morning.
It just runs.
This is the architectural distinction we kept returning to throughout our research and it's the thing the founders' neurosymbolic background, which we covered in Part II, most directly informed. For businesses serious about AI Adoption, this distinction is not a minor detail, it is the whole argument.
The Market Numbers Driving AI Adoption
The UK AI workforce automation market is currently valued at $4.4 billion. It is projected to reach $9.3 billion by 2031, growing at a compound annual rate of 16.1%. By 2033, AI agent revenue in the UK specifically is forecast to hit $9.67 billion.
Those are the headline figures. The more telling numbers sit underneath them. AI Adoption among UK SMEs climbed to 54% in early 2026 - up from just 25% two years prior. Among high-growth businesses, adoption sits at 98%. Eighty-nine percent of high-growth SMEs are already paying for AI tools. The market isn't forming. It has formed. The question now is which platforms capture it and as we established in Part I, Codewords is targeting three very distinct types of customer to do so.
The agile SME - ten to fifty people, losing hours every week to tasks like lead generation, content drafting, and inbox management, with no engineering resource to automate them.
The professional agency - small technical teams building AI Workflow Automation for clients at scale, running hundreds of thousands of agent tasks per month. And the regulated enterprise - larger firms in finance, legal, or healthcare that need automation they can audit, defend to a compliance officer, and run without surrendering data sovereignty to a US-based platform.
The crowded room they're walking into
Codewords is not entering an empty room.
Zapier, whose own people backed this round, as we noted in Part II, has 8,000+ integrations and a decade of institutional trust.
Lindy AI has carved out a polished niche in email and calendar automation with HIPAA and SOC 2 compliance.
Relevance AI targets engineering and data teams with multi-model flexibility.
SmythOS offers a full visual development environment with on-premise hosting.
Peak.ai is a UK-based specialist charging £12,500 per month for enterprise decision intelligence.
What Codewords has that most of them don't is the combination of natural language input and durable code output - the direct product of the research background we explored in Part II. You describe it in plain English. It runs as proper logic.
That distinction matters at the complex end of the use-case spectrum and that's precisely the end Codewords is targeting. For UK SMEs especially, that combination of accessibility and power sits at the heart of the AI Adoption problem the whole industry is trying to solve.
There's also a regulatory angle worth taking seriously. The UK's Data Use and Access Act 2025 made automated decision-making meaningfully easier to operate within the UK than under EU law. Codewords' London-based identity, its emphasis on bounded and auditable AI agent behaviour, and its zero-egress data positioning all place it well for UK enterprises that won't consider a US-centric platform. We believe that's a genuine structural advantage and not a marketing line.
"The bottleneck in AI doption isn't the AI. It's the interface." — Our assessment of Codewords
Where it could break
Reliability is the central unresolved question and it's the one that no funding round, no matter how well-backed, can answer in advance. Sixty-five percent of UK SMEs cite accuracy as their top barrier to AI Adoption. One AI agent that sends the wrong email, makes an unauthorised purchase, or deletes the wrong file in a live environment can set adoption back months inside a business. As the newer entrant, Codewords has to disprove that concern more urgently than any incumbent does.
The pricing model is the second thing we're watching. Credits roll over - a thoughtful design choice that removes the use-it-or-lose-it resentment that kills SaaS retention.
But once agents run at scale, tasks multiply fast and costs become hard to forecast.
Businesses that can't predict their spend tend to pull back. That's a known failure mode, and one the founders will need an answer to as they move upmarket.
And then there's the platform risk question that sits quietly underneath all of it: why won't Microsoft Copilot, Claude, or Google Gemini simply build what Codewords does natively within twelve months? We don't have a clean answer.
We're not sure Codewords does either, yet.
What we do know, having spent time understanding both the founders and the product across this series, is that the people building it have already walked away from one comfortable answer in favour of a harder, more honest question.
That instinct tends to produce better companies than the ones that never ask it.
The last piece of the puzzle - for now
Here is what we believe, having spent three parts examining this company from every angle we could find.
Codewords has the right architectural insight - that the bottleneck in AI Adoption is the interface, not the AI itself.
It has founders who earned their technical credibility before they went looking for commercial application.
It has a funding round that reads, in the calibre of its participants, as a vote of genuine conviction rather than trend-chasing capital.
It also has the things that €7.6 million cannot buy: the proof that comes only from running at scale, in live environments, under the kind of pressure that reveals what a system is actually made of. That proof doesn't exist yet. It's being accumulated, quietly, one automated task at a time.
We started this series asking whether a London startup that lets you type a sentence and have an AI agent run with it deserves the attention and capital it's received.
Three parts later, we think the honest answer is: probably yes.
But this story isn't finished. The market is moving too fast, the questions are too open, and the company is too early for anyone - including us - to write the final word. What happens next will be shaped by whether AI Workflow Automation at this level of accessibility can actually hold up under the pressure of real business use and whether AI Adoption at scale proves Codewords right.
Consider this series the beginning of a longer conversation.
We'll be back when there's more to say.
