Koru Kids built with
At the start of 2026, Rachel was using basic ChatGPT every day
Rachel set up Koru Kids ten years ago, and calls herself an early adopter. Before the Agent Accelerator she was already using ChatGPT “really extensively as chat”, as “basically a souped-up Google”, and nudging her team to do the same. Rachel was chatting with AI, instead of delegating work to agents.
In February 2026 she signed Koru Kids up for Cohort 1, about a week before it started, and went through it alongside four of her leaders. She moved from ChatGPT to Claude Cowork that same week.
Rachel sent four trailblazers from across the business through Cohort 1 with her
She chose leaders from operations, growth marketing and finance, plus the COO, who were already keen.
“I chose people who I thought would go for it and really embrace it. And they have. But also who I thought would become exemplars in their areas of business, and would inspire other people, and were drawn from different places in the organisation, so that it would spread virally.”
Koru’s leaders built working tools during the four weeks of Cohort 1
Operations Director
A timesheet app built to replace Deputy
Contractors log hours, managers approve them, and bills go straight into Xero. Google sign-in, an audit log and rates per area. The first working version came out of an 80-minute live workshop. It won the most creative prize at demo day, and a Koru team was testing it by May.
An estimated £1,000 to £1,200 a year if Koru moves off Deputy
Operations Director
An application to a government department
A policy-heavy application, drafted in Claude Cowork, then checked by hand.
Three and a bit days of work done in under three hours
Operations Director
A weekly plan posted to Slack every Friday
At 3pm on Fridays a scheduled task reads the week’s meeting actions and the calendar, then sends next week’s plan.
A scheduled task in Cowork, set up during the course
Finance Lead
A unit-economics dashboard
Lifetime value, acquisition cost and cohort views across both sides of the marketplace, built in Claude Code.
Stalled because the data pipelines weren’t ready for it
Four weeks in, Rachel called the course one of the best decisions she had made in a very long time
She recorded this on her phone the week the course ended. Signing up a week before it started had felt like a gamble: “it’s a bit of a punt, it’s quite a lot of time, you never know how your people are gonna take to these things.”
In four weeks, she said, her team had gone “from never really having done much of this stuff at all” to:
- “building their own apps”
- “replacing parts of what we do”
- “having so many ideas for things that they wanna do in the future”
- “having a really good understanding of security”
“The results have just been astounding, like transformational.”
She sent the four together on purpose: “putting four of them through the course at once has been really transformational actually, because we’d been our own little mini cohort.” She called them “the seed of a transformational change that is gonna radiate out through the company”.
She had worried about the mix of people. A couple of her leaders were more technical than the rest: “I was slightly worried at the start that it was gonna be boring for them, but it accelerated very quickly, brought everyone on board, and it’s just been such, such, such a positive journey.”
After the course, Rachel moved to Claude Code and now runs her work from a second brain on her own laptop
Soon after the course she moved to Claude Code, mainly because “about 10 different people” kept telling her it was more powerful. “As soon as I did it, it’s correct, is more powerful,” she said. “And it wasn’t as difficult as it might have been.”
- MarchSwitches from ChatGPT to Claude as the course starts
- During the courseBuilds her first small tools, like a monitor of UK parenting threads on Reddit
- After the courseMoves to Claude Code, then builds a second brain on her own laptop
- OctoberBuilding a hiring hub and Koru’s own applicant tracking system
Then she built a second brain: folders of notes on her laptop rather than inside any one AI product, so she can switch models without starting again. “I’m really, really confident that the way that I’ve built it, I’ll be able to just plug anything in.” It holds thousands of files, including more than 2,000 on people she knows. Voice notes go in from a button on her phone’s lock screen and are transcribed on her own machine, so they cost nothing.
Granola, which she turned on for every meeting and connected to Claude during the course, now feeds her one-to-one skill. It reads the latest notes with that person, what they are working on and her own priorities, then suggests questions. A scheduled task looks through Slack for two or three things a day she can thank people for. And when someone she met months ago comes back, Claude preps the call: “my questions are 10 times better, it reminds me what the sticking point was last time.”
She keeps her inbox away from all of it. After the course covered prompt injection she did not connect her work or personal Gmail to any AI, and forwards whatever she wants Claude to handle to a separate AgentMail inbox instead.
Koru brought along the people further back with all-hands demos and a weekly AI tip
Rachel’s aim is to take a ten-year-old company “as close to being AI-first as we can”, and she says no single tactic does it: “different kinds of people respond to different things in different ways.” Beyond the trailblazers, she asked people who were further behind to demo at all-hands, “because what I found is it’s powerful to have someone who everyone knows is not one of those people actually do a demo.”
Every all-hands now has an AI tip of the week. Rachel alternates them between the cutting edge and the basics, for colleagues who are still “totally switched off and bamboozled” by talk of new models.
In May, Koru’s Operations Manager joined Cohort 2. On the first call they said, “I’m not anti-AI, but I am definitely on the more sceptical side”, and wary of relying on it too much. Within weeks they had vibe-coded their own task app. A daily briefing now reads their calendar and tasks and posts to Slack, and a skill turns yesterday’s meeting notes into tasks.
A shared skills library means every Claude at Koru starts from the company’s own context
By May, Koru had an organisation-wide skills library inside Claude, looked after by an admin, with people switching on the skills they need. It held:
- Looker and BigQuery, so everyone’s Claude asks the data warehouse the same way
- Brand guidelines, so internal apps look like Koru
- Insights by customer group
- Liquid, the templating language of their email platform
- Ask Koru Kids, the customer help content, through Intercom
Koru’s rules for vibe coding live in that library as a skill, so every build starts from the same approved tools and the same idea of what counts as launched.
Now Rachel is moving the company’s knowledge into Notion, which she calls the third brain, one problem at a time. “We’re expecting it to be a long, slow, iterative transition, rather than a huge project where we down tools and everyone builds Notion.”
Six months on, Koru has moved AI from side projects into the core of the business
“The next stage of evolution, I think, has been from little toy things, which were what we were practising while we were on the course, and things that were incidental to the way we work, to things that are much more fundamental and intertwined with our core business.”
While learning: small and incidental
- A monitor of parenting threads on Reddit
- A personal task app and daily briefing
- A one-off application to a government department
- A weekly plan posted to Slack
Now: core to how Koru runs
- A hiring hub and Koru’s own applicant tracking systemHow Koru recruits, rebuilt in-house by the CEO. CEO, in progress
- BigQuery and Looker connected to ClaudeAnyone can question the company’s data in plain English, without knowing the query syntax. Company-wide
With BigQuery and Looker connected to Claude, “that means that I, as a CEO, just have so much more ability to understand what’s actually going on in my business,” Rachel said. Whoever runs a marketing channel can now follow conversions much further down the funnel and target spend better: “that is core, core, core business functionality.”
The night before her October Lunch & Learn, Rachel built a hiring test for sales candidates
Koru needed one or two more people for a three-person sales team, and AI-written applications had made the pile hard to cut. “Now you can weed out like 10 of them, and 190 of them kind of look credible.” So Rachel built a test that sounds like the job itself. A candidate presses start, gets five minutes on a timer, hears a voice note from a parent and rings back to leave a voicemail. The team’s side sits behind a Koru Kids Google login, with each recording, the brief and a marking grid.
Her sales hiring manager asked how they would ever get it built. “I said, do not worry, I can do it.” She built it in an evening with Claude and Twilio, reusing the phone plumbing from an earlier experiment.
She closed the Lunch & Learn with a word for anyone who had never heard of those tools:
“If you’re kind of thinking, well, that sounds cool, but I don’t know what Twilio is, I don’t know what Netlify is… obviously, neither did I. Claude will just tell you.”
Every new hire at Koru now has to raise the company’s average AI ability
Rachel changed her mind about who should build. “I had this thought a few months ago that, oh great, it used to be that only some people can build, now everyone can build, therefore everyone should build. And I think I’ve really realised that not everyone wants to build. And that’s okay.”
She holds one line: “What we can’t have is anyone opposed to AI.” Every Koru hiring process now has a step that checks for it, and the bar she has set the company is that “everyone who we bring in needs to increase the average ability of the company in terms of AI.”
Rachel calls the Agent Accelerator one of the highest-value things Koru could have done
By October she was rebuilding how Koru hires, and she put that down to the course: “I definitely wouldn’t have the confidence to do that if I hadn’t done the course and then built various things afterwards.” In the video she recorded the week the course ended, her verdict was already clear:
“I think it is one of the highest value things I can possibly think of that we could have done. And I think the course itself is just really well-designed.”
Rachel Carrell, CEO & Co-Founder of Koru Kids, the week Cohort 1 ended
Start with your own trailblazers
Rachel and four keen leaders went through a live Agent Accelerator cohort first, then spread what they learnt with all-hands demos and shared skills. Cohort 5 starts Tuesday 17 November 2026. To train a whole team at once, James also runs private workshops. You can read 119 reviews from the people who have done it.
Four weeks, eight live sessions, about 4.5 hours a week. See the price and the curriculum.
