42 Days with Claude Cowork: £0 Revenue, Real Work Lessons
42 days with Claude Cowork, £0 revenue, 9 roles. Measured real work created, not hype, and the lessons for solo founders.
42 days with Claude Cowork, £0 revenue, 9 roles. Measured real work created, not hype, and the lessons for solo founders.
Published on 27 March 2026
In business, revenue is a useful signal. It is also a blunt instrument. So when I ran my Claude Cowork partner experiment over 42 days, I refused to measure it the way most “AI success” stories do.
The headline is simple: I spent around £163 on Claude (subscription costs), I generated £0 revenue during the experiment window, and I still ended up with meaningful work created across multiple parts of my business. Not just drafts, not just ideas, but actual artefacts that now exist in folders, documents, templates, pipelines, and systems.
The point is not that revenue does not matter. The point is that revenue is often a second-order outcome. It depends on timing, distribution, product readiness, trust, lead flow, and conversion systems that are mostly outside the tool.
My first question was better: what would exist now that would not exist if I had done nothing for 42 days? That is the difference between measuring “AI productivity” and measuring “work performed”.
In this post I will share the numbers, the caveats, and the practical lessons for founders who live close to the marketing and execution ceiling. If you are trying to understand the bigger context for why work gets fragmented across apps and conversations, you might also want to read the 16 context islands problem.
Here are the headline figures from the experiment, in plain language.
£163 in subscription costs.
An estimated £54,015 in human equivalent value.
Roughly 116 working days compressed into about 14 hours of my time.
Plus, the “shape” of the work: across nine distinct categories (roles), Claude Cowork supported tasks that typically would require either hiring, subcontracting, or months of accumulation of partial progress.
That last part matters because a single task time-saved metric can hide what the tool is actually enabling. Cowork does not only accelerate one lane of work. It reduces the switching cost between lanes, which changes what you can realistically attempt in a single week.
I need to be honest about the numbers. The £54,015 estimate is interesting, but it is not a clean audit of value created.
It is based on an “equivalent professional wage” model, and it assumes you would have hired someone for each task I prompted. In reality, I would never have hired contractors for every small job. Most of that work would have remained on my “someday” list, not been outsourced.
So the real question is not “how much did I save?”. The real question is: what exists now that would not have existed without this experiment.
That is also why a tool cost can look small while the output looks large. For solo founders, “value” is often about finally completing work that you would otherwise delay indefinitely.
The second reason the experiment feels more meaningful than a simple productivity report is compounding.
Sessions 1 through 9 were mostly setup: building the tracking system, the knowledge architecture, and the session protocol. Those nine sessions created around £14,675 of equivalent work.
Then sessions 10 through 46 built on that foundation. The next 34 sessions produced around £39,340 of equivalent work, using the same tools and the same person.
So the work output increased without a proportional increase in marginal effort. That is a founder-friendly pattern: investment in systems early, then leverage later.
My tracking is not perfect. Claude was used through multiple surfaces, including Cursor coding work. Some sessions did not make it into the tracker, and a few Cowork sessions were missed when I forgot to run the end-of-session protocol.
Because of that, the experiment is likely north of 50 sessions, not exactly 46. Treat the numbers as a credible direction of travel, not a forensic ledger.
Also note that the experiment includes an overlap effect. I ran Cowork while Claude Code or terminal tasks ran in parallel. My time overlapped, which means the effective cost per session is lower than it would be if I had single-threaded everything.
Over 46 sessions, Claude Cowork worked across nine categories. Not because I planned it that way. Because that is what a solo founder’s week looks like when you ask one partner to help with whatever lands on your desk.
That breadth is the reason this experiment is useful to founders. It is not only “AI can write”. It is “AI can help you do the whole job as it actually happens”. Research, drafts, strategy, development tasks, IT infrastructure, design thinking, and coaching support all show up in a founder’s calendar at inconvenient times.
In my case, the switching cost dropped to nearly zero. For example, tasks that used to sit in a “someday” pile for months became achievable because the assistant could carry context within the workflow and help complete adjacent steps.
The practical headline is that the experiment cost was about £3.88 per day for a work partner that could span multiple roles. If you are a founder, that is not just a cost saving. It changes your default planning behaviour.
Some work would not have happened without the experiment. I would not have carved out time to do a video editing pipeline, a security audit, and a branded PDF template system on top of everything else.
The semantic search engine across my knowledge base may be the most valuable long-term asset. It changes how quickly I can find “answers” inside my own business history. That is a different kind of ROI than revenue today.
All of this points to a founder reality: many delayed tasks are delayed because the cost is not the execution time. The cost is the mental overhead of getting back into the problem context, then doing the work across multiple tools without dropping pieces.
Claude Cowork does not remove judgement. It removes the friction that prevents you from attempting the work at all.
The question you might be asking is simple: “If it did not make money, why does it matter?”.
My answer is that I was not trying to manufacture revenue inside the experiment window. I was testing whether a persistent AI partner could turn my “someday list” into real outputs.
Before the experiment, my weeks had a hard ceiling. For me, the practical capacity looked like:
Everything else competed for scraps. Infrastructure, systems, creative projects, and longer-term improvements were always at risk of being postponed until “later”. But later rarely came.
The experiment changed that by reducing the mental and financial cost of non-client work. A short session before a client call could produce a branded handout. A lunchtime session could index a thousand files. The work did not need a dedicated afternoon anymore. It needed direction, judgement, and a taste check.
It also improved client outcomes indirectly. When I prep for a discovery call now, I can search across my notes, previous conversations, and frameworks. I walk into calls with better context. Better questions follow. That means clients get more value from the same time slot.
So even though revenue was £0 during the window, quality and execution momentum improved in the places that typically determine revenue later.
Plenty of AI discussion focuses on adoption rates. Those matter, but founders do not need another dashboard that says people clicked something. Founders need a different measurement: does the tool help you create the artefacts your business requires, with a credible quality bar?
That is what I mean by “work ROI”. It is the stuff that exists after the session: systems, templates, pipelines, knowledge structures, and repeatable workflows.
If you want the dataset view of the same experiment, you can see the dashboard-style write-up at 40 Days with Claude Cowork: The Full Breakdown.
If this post reads like a win, I want to correct that. The experiment had real weaknesses. This section is the trust builder.
I asked Claude to build out a digital product portfolio and pitch it. The strategy was ambitious: prompt templates, a branded PDF bundle, a Gumroad page, PromptBase listings, and a marketing plan.
Claude did that all in one sitting and asked me to promote it. But marketplace listings without a warm audience received no organic traffic. So there were no sales inside the experiment window.
This is a reminder that tools do not create distribution. Distribution is a founder skill, and it needs time, repetition, and an outreach system.
Claude can carry context within a session. Between sessions, you are re-establishing habits. In my case, Claude forgot the end-of-session protocol four times, and I had to fix it four times.
Context needs a waypoint system. The next session should load a summary of key decisions and outcomes. Without that, it is like managing a smart contractor who forgets where they left off yesterday.
Some sessions were spent fixing Sovereign Brain crashes, including out-of-memory errors, corrupted indexes, and recovery procedures.
AI created parts of the system, then AI had to help debug the system. That overhead took my time, and it must be factored into any honest time accounting.
The “£54k saved” number assumes I would have hired someone for each task. I would not have.
Claude overestimates how long a human would take to do the same work. I spot-checked the numbers against a few invoices when values looked inflated. Even if the traditional equivalent is closer to £30k to £40k, the lesson remains: the value is in the work created, not in the precise number.
If you are going to repeat this experiment, do yourself a favour. Track what was produced and whether it would have existed otherwise. That is the metric you can actually defend.
Microsoft launched “Copilot Cowork” this month, and it validates the category I have been testing: persistent AI as a work partner rather than a one-shot tool.
When big platforms validate the behaviour, the direction of travel becomes clearer. You can read more in Microsoft’s announcement of Copilot Cowork.
The next phase for me is not only time savings. It is generating revenue through autonomous workflows that can run without me being constantly in the loop.
In other words, new products and distribution assets built from persistent context. I want to focus on client work while an agent handles repetitive production loops, with checkpoints that I can audit quickly.
That is why persistent context matters. The fewer times I need to re-explain my business, the more I can push the machine from “drafting” into “execution”.
If you want to run a similar experiment, you do not need a giant tool stack. You need a simple workflow that answers four questions:
This aligns with the broader idea behind making your business knowledge reachable. It also pairs naturally with strategy work, where the goal is not “more prompts”, it is “better decisions and better distribution”. If you want help turning AI experiments into a plan, our marketing strategy service is designed for exactly that founder problem.
The ultimate value of this 42-day experiment is not the £54,015 estimate. The real value is what now exists in my business that would not have existed otherwise.
I now have a knowledge system, a video editing pipeline, a security-aware workflow, a streamlined coaching reflection practice, a consistent branded PDF creation path, and a habit of publishing with less friction.
For solo founders, that matters because your “someday list” is where real business value hides. The amazing part is that the execution cost has effectively dropped to near zero. The blocker is education and knowing what is possible.
If you want to go one step further, reply with your “someday list” and what you want to unlock this week. I reply to everyone. And if you would rather turn it into a structured plan, you can book a Discovery Diagnostic.
Explore more insights and strategies to elevate your marketing approach.
Honest results from a 42-day Claude Cowork experiment. 46 sessions, £163 actual cost vs £57,593 traditional equivalent. Every session, every role, every number — the full dashboard.
Claude Skills turn organisational knowledge into reusable AI modules. For founders of £1-7m businesses, this means multiplying productivity without scaling headcount. Here is why it matters.
A hands-on guide to building Claude Skills that remember your brand, audience, and standards. Real examples from three months of daily use in a founder-led consultancy.