How a solo consultant uses an AI operating system to multiply his output.

Nick was already a heavy AI user before this. He ran his work through ChatGPT and Claude Projects, one project per topic, each loaded with instructions and examples by hand.

• He curated context once, and it decayed. His LinkedIn project held his tone-of-voice rules and favourite posts, uploaded when he set it up, and two years later it was still writing from those two-year-old references, because after every good post he was never going to copy it out and re-upload it.
• He fixed outputs instead of instructions. Conversations ran "no, that's not what I meant, do it like that, I also explained this last time", and the same corrections repeated week after week.
• The projects couldn't share what they knew. His course frameworks feed his consulting, and his consulting produces the stories that improve the course and his posts, but Claude Projects are siloed by design, so none of that context could flow.
• The CRM lived in his head. Notes on what a prospect cared about existed only for the biggest stakeholders, because keeping them for everyone was tedious, and with sales cycles of 6 to 18 months, details were long gone by the time a prospect came back.
• Transcripts were an occasional export. He used an AI notetaker, but only pulled transcripts into Claude for the big meetings, so most calls left nothing behind.

The deeper cost sat underneath the repeated corrections. Whole categories of work made no sense to hand to AI at all, because fixing the result took longer than doing the job.

Click to expand

A slide from Nick's actual presentation on the system, telling the before story. Knowledge scattered across tools, AI as a thought partner only, context pasted in by hand, and the question that set up the rebuild.Close

Claude Code came out and Nick picked it up, in his words, very aggressively, in February 2026. A decade working alongside data scientists and engineers gave him a frame for using a non-deterministic tool.

His frame is that you can't test such a system into being always right, you work out the cost of it being wrong and design around that.

"It feels crazy that I've not worked with it my whole life now," he says.

The core business context lives in one repo of markdown files, which he views through Obsidian and works on through Claude Code or Codex. A global CLAUDE.md file sets the rules, and cascades down into more specific ones.

Click to expand

Demo screenshot with dummy data. The whole business in one sidebar, CRM, tasks, meetings, projects and a review queue, every one a folder of plain markdown.Close

• Context stays fresh on its own. Every call transcript syncs into the repo automatically through Granola's Obsidian plugin, and rules tell the AI what to update after each call. Nothing waits for him to curate it.
• Instructions get fixed, not outputs. When something goes wrong, he asks what rule needs to change so it never happens again, and the rule gets rewritten. "When a mistake happens, we fix the root cause. We don't just fix the symptom."
• Everything shares one brain, with walls where they matter. Each client or internal project has a folder with its own CLAUDE.md, so he can say "I just got off the phone, update the proposal accordingly" and it knows exactly who and what he means, while work for one client can't leak into another's documents.

Click to expand

Demo screenshot with dummy data. A project's CLAUDE.md, holding client, contact, deliverables and conventions, which is why "update the proposal" needs no further explanation.Close

• The CRM writes itself. A folder holds a markdown note per company, person and opportunity, and he never edits them by hand. After every call the log updates, and the template captures what each person cares about and the metric they're judged on, so a proposal for a marketing lead talks conversion rates and one for a finance lead talks profit. When a prospect resurfaces after six months, he asks for the exact words they used to describe their problem, and writes the next email in them.
• Tasks write themselves too. Each task is a markdown note the AI creates, with status, priority, due date, an AI-estimated duration and links to the relevant files, under a standing rule that anything the AI can start, it starts. If he promises an introduction on a call, the intro is drafted before he asks.

Click to expand

Demo screenshot with dummy data. A task the AI created for itself, with status, priority, due date, an effort estimate and links back to the note it came from. He never writes these by hand.Close

• He approves anything that leaves the building. Drafted emails land in his Gmail drafts folder, and he reads and presses send. For simple introductions, where the CRM knows both people, that review takes a glance.
• The day starts with a brief. The AI writes a daily brief of top priorities, important emails and anything overdue, and a shorter version arrives as a Telegram message each morning. He has ADHD and a billion tabs open at any moment, he says, and the brief is what pulls him back to the thing that matters.

Click to expand

Demo screenshot with dummy data. The daily brief, with priorities, today's tasks and follow-ups compiled overnight from the notes and transcripts underneath.Close

Click to expand

Demo screenshot with dummy data. A CRM person note, with role, relationship and an interaction log that updates after every call without him touching it.Close

The whole vault syncs through Obsidian Sync and Git. Git earns its place because it lets the AI trace exactly what changed, roll anything back, and run parallel agents on different parts of the vault without them clashing.

The files are the system, and what has changed over the year is how he looks at them.

Everything above is what he calls the initial version, which was Obsidian and nothing else, every note read and edited in the app. The screenshots so far are that layer, and he still uses Obsidian, just less than he did.

The shift started with one-off requests. He would ask for a plan as an HTML page rather than a wall of text, because it was easier to take in at a glance.

Those one-off pages became permanent. He now has a local web app that starts up with his laptop, with a module for each part of the business, and it is where he reads most of what the system produces.

• The dashboard is the daily brief made visual. Top priorities, flagged email, anything overdue, and the objective for the current cycle.
• The CRM module shows the pipeline, what's in flight, what needs a proposal and the total value, all read from the markdown notes underneath.
• The community views are why he didn't buy a CRM. He says he'd have paid for HubSpot if a pipeline was all he needed, but he also tracks webinar guests, meetup sponsors and guest article contributions, so he had the AI build views for those instead.
• A demo mode swaps in randomised data, which took about five minutes to build, and is what let him walk us through all of this without showing a single client's details.

Click to expand

Dashboard shown with dummy data: priorities, flagged email and the objective for the example cycle. The tasks and messages are illustrative, not actual client work.Close

Click to expand

CRM pipeline shown with dummy data. The names, opportunities and deal values are illustrative. Webinar speakers, meetups, sponsors and guest articles get their own tabs, which is why an off-the-shelf product didn't fit.Close

The app reads the same markdown, with a local SQL database for data that doesn't suit notes, like his synced bank transactions. AI-built code powers the dashboard and integrations.

Running a local server sounds like a thing that breaks, and in the first month it did, with the app shutting itself down after periods of inactivity. He fixed the cause rather than restarting it each time, moving it to a Launch Agent that keeps it running.

Build time. There was no big build. He started in February 2026 and has added to it continuously since, module by module, keeping what earns its place.

His test is blunt, "adoption is the real test of whether something is good," and anything he stops using gets changed or killed, because unused modules are bloat that makes the rest worse. We've asked for his estimate of the total hours invested.

Difficulty, hard. Months of iteration, and you have to enjoy it. He plainly does.

Prerequisites. A Claude Code or Codex subscription, comfort living in plain-text files, and one habit that matters more than any tool, treating every annoyance as a broken rule to fix rather than a bad output to correct. His decade working alongside data and AI teams shaped the mindset, and the foundation itself is markdown notes and plain-language rules, with AI-built code for the dashboard and integrations.

The figures below are Nick's own, sent after the interview. The percentages are his estimates rather than tracked measurements, and he says so.

• Average monthly revenue in the first half of 2026 ran at 2.4 times the 2025 monthly average. That is from his own revenue figures.
• On one engagement he estimates AI cut delivery effort by about 45%. The system kept track of a large amount of stakeholder and project information at a level of detail he says he would have struggled to sustain otherwise.
• He could turn stakeholder feedback into an updated prototype about two hours later, built in Lovable. The combination of that detail and that speed has produced referrals and follow-on work, he says.

Click to expand

Real colleague feedback, with names redacted. From the engagement where he estimates AI cut delivery effort by about 45%.Close

Click to expand

Real colleague feedback, with names redacted. More from the same thread, where the eye for detail is his and the system is what let him sustain it across the engagement.Close

Where the hours went. These are his estimates of hands-on time, before and after.

Speed matters more to him than the hours. Proposals now go out within zero to two days instead of one to two weeks, which means reaching an opportunity while it is still warm. Events get published sooner too, so guests have more notice and more time to sign up.

• He puts £200,000 to £300,000 of pipeline down to the system. That is AI upskilling and operating-model work he says arose specifically through the personal OS and the conversations it started, on top of a decade of enterprise data and AI experience.
• Tasks that weren't worth delegating now are. The old maths, where fixing AI's work took longer than doing the job, has flipped, he says.
• 99% of his work happens in one interface, his estimate, instead of a round trip through a design tool, a chat window and half a dozen tabs.

The revenue picture. The 2.4 times figure is real and his own, and he is careful about what it proves. Some of it is a young business growing, he says, and some of it reflects the extra capacity and quality he has been able to deliver.

Every webinar is hosted on Circle, his community platform, and promoted through Luma, where his meetup mailing lists live. Publishing one meant a chain of small manual jobs.

• He made a landscape banner in Canva at Circle's required dimensions.
• He made a separate square image because Luma's format is different.
• He wrote the Circle description to the community's usual format.
• He rewrote it for Luma, where readers might not know the community, so it needed joining instructions and an explanation of what the community is.
• He checked every link and published the event to the London, Barcelona and Paris calendars, one by one.

The tedium was half the problem, and sequence was the other half, because none of it could start until the guest confirmed their photo, bio and title. He'd set aside a Thursday to prepare webinars and get stuck waiting, and by the time a guest confirmed he'd be busy with client work, so invitations went out last minute and too few people joined.

Once the personal OS existed, the webinar process was a folder of written SOPs away from automating. The AI already knew every guest from the CRM, and every step was a rule it could follow.

Now, when a guest confirms their title, he tells the system it has everything it needs. It generates the banner and the square image, drafts both descriptions, and waits.

He checks one thing, the banner, mostly that the guest's photo looks right. Then he says publish, and the event goes up on Circle and on all three Luma calendars, with the right links.

The connection runs through the Luma and Circle APIs, with one exception. Luma's API wouldn't update the event thumbnail, which used to be a manual step he kept for himself.

Now the AI does that step through the browser, a small, self-contained task he trusts it to do and, in his words, not mess up.

The same image pipeline handled a rebrand. He had the AI regenerate every old webinar thumbnail in the new design, reviewed the set, gave feedback on the one that looked wrong, and then had it update the whole library, a job he says he would simply never have done by hand.

Click to expand

An actual published event page, with banner, description and listing all generated. His ten-second look at the banner is the only human step left.Close

Click to expand

The actual community recordings library after the rebrand. Every thumbnail was regenerated to the new design in one request, then updated across the library.Close

Build time. Not asked on the call, and we've followed up. The shape of it is a written SOP, two API connections and iteration on the image prompts.

Difficulty, medium. A few weekends, comfortable editing text files and breaking things. The API connections are the technical part, and browser use now offers a slower but simpler route where an API is missing.

Prerequisites. Accounts on your event platforms, a written SOP of your current manual process, and enough stored context about guests for descriptions to come out right, which his CRM provides.

• An event banner and its descriptions take about 5 minutes of his time, down from about 40. That is his estimate, and at his planned volume he puts it at roughly 25 hours a year.
• The community is on course for around 33 virtual events this year, against 8 in 2025. He says he could not sustain that programme alongside client work without the automation.
• The dependency trap is gone. Publishing happens the moment a guest confirms rather than whenever he next had a free Thursday, so invitations go out with more notice and guests have longer to sign up.