{"componentChunkName":"component---src-templates-blog-template-js","path":"/an-agentic-refounding","result":{"data":{"contentfulBlogPost":{"title":"An Agentic Refounding","featuredImage":{"fluid":{"src":"//images.ctfassets.net/sg5rvwjajl1u/5z5h2wlxW7EI9ixR03ALlC/cc64072e6d4a917d5e0aa030b5889256/agentic-refounding-square-1200.png?w=800&q=50"}},"date":"3rd November, 2026","body":{"json":{"data":{},"content":[{"data":{},"content":[{"data":{},"marks":[],"value":"Two thought pieces have done the rounds this year that I keep coming back to. In February, Toby Mather published \"","nodeType":"text"},{"data":{"uri":"https://deliberateceo.substack.com/p/opus-babies"},"content":[{"data":{},"marks":[{"type":"underline"}],"value":"Opus Babies","nodeType":"text"}],"nodeType":"hyperlink"},{"data":{},"marks":[],"value":"\", arguing that companies founded after Anthropic's Opus 4.5 landed last November are a new species of organisation, where one person and a team of agents are doing the work that used to need a whole squad. Then in March, Jack Dorsey and Sequoia's Roelof Botha published \"","nodeType":"text"},{"data":{"uri":"https://block.xyz/inside/from-hierarchy-to-intelligence"},"content":[{"data":{},"marks":[{"type":"underline"}],"value":"From Hierarchy to Intelligence","nodeType":"text"}],"nodeType":"hyperlink"},{"data":{},"marks":[],"value":"\", which, through two thousand years of organisational history, explained how hierarchy only ever existed to route information between humans, and that, now AI can do that job, a new model of organisation is required.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"The authors begin from very different starting points: Toby's new company, Rig, started from day zero on a completely clean sheet of paper, whereas Block is a publicly listed company with huge distribution and an established business model. But for all the visible differences, in some ways they both have enormous advantages: newly founded 'Opus (or now Fable) Babies' are designed around agents from the first line of code, and Block can afford to rebuild while the existing machine keeps paying for it.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"But what about the companies in the middle? Those less than a decade in, VC-backed, with embedded ways of working, but without the scale or coffers of a large public company. This is exactly where Sidekick sits, and I suspect it is where most of the fintechs founded in the 2015 to 2022 wave sit too. It's a hard truth, but an Opus Baby can now build most of the app features we have spent three years building in a fraction of the time, while the bigger players have the budget to throw at the problem. The middle has neither the blank sheet nor the balance sheet. It's the most uncomfortable place of all to be tackling the AI question from…","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"So this is the first of a series, part diary of a rewire while it happens, and part attempt to open up a conversation with others on the same journey, because I know we're not the only ones working this out in real time.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"To jump to the punchline, here is what I've learned so far","nodeType":"text"}],"nodeType":"heading-3"},{"data":{},"content":[{"data":{},"marks":[],"value":"There is no half-rewire. You can't bolt agents onto a company designed for humans and end up with an agentic company. One converted founder inside an unconverted company changes nothing. The company has to convert, top to bottom, while the plane is still flying, and that is a bigger ask than it sounds, because the notion of a company is embedded in human behaviour and human structures that have existed for a very long time. Dorsey's history lesson is right: every layer of management, every department, and every sign-off chain is an information-routing system built around the fact that one person can only coordinate a handful of others. But bring agentic, AI-first ways of working into that and the whole notion of what a company function or department is gets thrown up in the air. Dorsey has a ","nodeType":"text"},{"data":{"uri":"https://sequoiacap.com/podcast/jack-dorsey-every-company-can-now-be-a-mini-agi"},"content":[{"data":{},"marks":[{"type":"underline"}],"value":"phrase","nodeType":"text"}],"nodeType":"hyperlink"},{"data":{},"marks":[],"value":" for what follows that explains it best: \"companies have multiple founding moments\". ","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"This is one of them.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"The Google Translate year","nodeType":"text"}],"nodeType":"heading-3"},{"data":{},"content":[{"data":{},"marks":[],"value":"Don't get me wrong, we weren't ignoring AI in 2025. If anything, we thought we were keeping up with it. Every tool we used that had an AI feature, we switched on; Midjourney was producing our illustrations; our engineers were using Cursor. And we shipped some low-risk AI features in the Sidekick app, such as matching customers' support questions to answers we'd already written, because we didn't yet trust that the models wouldn't hallucinate. We also reached the final of the PIMFA AI competition with a chat-based fact-find that was, in substance, a thin layer over ChatGPT (which in itself is less a confession than an indication of how fast this has moved). I was knocking together internal tools and product prototypes on Lovable at weekends, and my eyes were starting to open.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"All of it was learning and all of it was additive. But it was the Google Translate approach to speaking a language. You carry on doing things the way you've always done them, and when you hit a part you can't do, you paste it into the box and a decent answer comes back. It helps, but it doesn't make you fluent, and it doesn't change what you are.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"In 2025, AI felt like a technology to help us do what we do today faster, better and more personalised. ","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"In 2026, I couldn't see things more differently.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"So what changed?","nodeType":"text"}],"nodeType":"heading-3"},{"data":{},"content":[{"data":{},"marks":[],"value":"The first quarter of this year did, and it was an accelerating, enthusiastic 'red pill' moment rather than a slow dawning. Starting with my own job, anything that regularly landed on my plate as a repeatable task, I tried to hand off to Claude: portfolio monitoring, our weekly market commentary, budget forecasting, the board pack. I gave myself a rule that if a task was going to take an hour, I would spend the hour automating it instead so that I never had to do it again (a rule I broke constantly, although the direction held), and within weeks work that had filled my evenings was running end to end without me. These were strong personal proof points, and other team members were starting to have similar gains too, for instance our Operations team running customer campaigns, payroll and business planning in a similarly automated way.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"Crucially, though, those early attempts kept the human structures and simply put Claude inside them: driving Figma through Claude with guardrails around it, getting Claude to populate an existing Google Slides board template, using the existing tools and processes with an agent in the seat. It was a halfway house and more of a sticking plaster over a root cause: that essentially every one of those tools and processes only existed because a human had to do the work. Once you can automate everything, you discover that everything underneath was built in the image of human-shaped work. But for agent-shaped work, it has to be rearchitected. Context and coordination are king, and neither existed in a form an agent could use, because our data sat behind human-gated logins, and our knowledge sat in the files (or worse, heads) of the people who did the work, inaccessible or unnavigable for the wider company.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"The step change started when we got away from the human tool and the human process altogether and asked what the work would look like if it had never been designed for a person.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"Every working day since, and even the long nights during my paternity leave, I've been asking the same questions. What does this mean for us as a company? What does it mean for me personally as a founder? How do I challenge every assumption I have, get to ground zero and build back up correctly?","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"Not an iteration, a reimagining","nodeType":"text"}],"nodeType":"heading-3"},{"data":{},"content":[{"data":{},"marks":[],"value":"Looking back, I got some things right at the start. I made sure I banged the drum whenever I could, and I led by doing: building a proper second brain for my own work, prototyping the operational tools myself, and working closely with one of our engineers who was an evangelist on all of this too, until the conversation began to spread more widely.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"However, you can increment and increment and never become an AI-first company, because an AI-first company thinks differently. By the middle of the year it was clear that change had to come in more of a big bang than by incremental dripping, and it was equally clear, looking at the size of the opportunity in front of us and the time we had to go after it, that the operating model itself had to change.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"It's never a nice thing to do, but we changed the shape of the company. That meant unfortunately losing some good people (the worst bit of any Founder’s job), but it was also the chance for real, profound change. In practice, the old 'company functions' have given way to people who own outcomes and build with agents to deliver them; engineers increasingly architect and review rather than write every line themselves; compliance sits inside the build rather than at the end of it; and the knowledge base, not a management layer, is where the context lives. With a smaller team and the collapsing of traditional departmental structures, the tone going forward has been completely different: we are all builders now; engineers are no longer the only people who ship; and how the company enables everyone to build safely, with agents at the centre of every process and product, becomes the requirement.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"What a refounding actually consists of","nodeType":"text"}],"nodeType":"heading-3"},{"data":{},"content":[{"data":{},"marks":[],"value":"A refounding is not a list of tools. The foundations of the business had to be reconsidered, and that is where a significant part of my headspace this year has gone.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"Firstly, we have centralised the company's knowledge and pulled it into a compounding feedback loop: how we work, what we've decided and why, what the data means and what the lessons are from the past. Constantly updated, constantly improving, constantly available. It gives everybody (humans and agents) the context and the access to information they need for their work, and it has been the single greatest unlock. It has enabled the parallelisation of work, a step change in velocity, and a significant reduction in overhead and time spent on coordination and planning.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"Secondly, safety and security had to be rethought from the ground up, because if an agent can be involved in the work, that raises immediate questions about access, privacy and what a thing that can run on its own may touch. The answer has been the right air gaps and tooling to let more autonomous, parallel, multi-agent teams work on real problems, while we stay comfortable that they can't step outside their mandate.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"Thirdly, the way things get built has inverted. Everybody in the company can now prove an idea with a real, working prototype rather than a requirements document or a Figma mock-up, and once it's proven, an agent team can pick it up and productionise it. The biggest test of that so far is a whole new business line on the horizon, which I'll write about properly very soon: its first working version was proven exactly that way, as a prototype, before an agent team took it on, and because it sits separately from the existing Sidekick core business, we've been able to be even more progressive in how agent-first we've made it. What we learn there gets pulled back into the core.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"The board pack is the smallest example and still the one I enjoy most. In March I was trying to force an agent into a slide template; today, because so much context sits underneath it, the first version arrives at the standard of a strategy consultant, and my job is final 'partner review' rather than moving boxes around a page or hunting for the right data point. Our next iteration will be an MCP that board members can query directly via their own agents to get ringfenced but relevant context for board specific matters.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"Who's accountable when the agent did the work?","nodeType":"text"}],"nodeType":"heading-3"},{"data":{},"content":[{"data":{},"marks":[],"value":"We don't have six thousand people and multiple layers of management to flatten, but we are a regulated entity. We manage client funds, and trust is everything. You can't simply burn down the structures of a regulated entity and see what grows back, which is why agent accountability is THE question for a regulated business, and the regulator is asking it too. The FCA's position is that its existing framework applies, with the senior managers regime at the heart of it: senior managers are personally accountable for the AI systems in their area of responsibility. Tellingly, its own review of good and poor practice holds up accountability assigned from the design stage as good practice, and found that retrofitting governance after deployment \"proved difficult and left gaps in accountability\".","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"This chimes a lot with how we've been thinking and designing through this 'refounding'… A human stays the accountable signatory, and we design so that an agent can never become one. What changes is what sits underneath. A compliance function at a traditional firm our size would historically have relied on dip tests, reviews on a fixed schedule, and annual training programmes as the crux of their control environment. But now, with AI, the same named person can see more than ever, with the context attached, and the record keeps itself, which is a stronger control with a better audit trail rather than a weaker one. And when the compliance rules sit in the context the agents build from, the feature, the marketing draft or the customer email arrives for sign-off mostly right, rather than landing on an approver's desk the day before launch with lots of rework required.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"The half-rewire fallacy","nodeType":"text"}],"nodeType":"heading-3"},{"data":{},"content":[{"data":{},"marks":[],"value":"The obvious objection is the one some market analysts are now putting out there: for all the hype, where is the fundamental business improvement from AI showing in the financial statements?","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"An important part of the answer, in my mind, is the half-rewire fallacy. The appearance of AI productivity with the 'Google Translate' approach looks exciting from the inside: nothing has broken and simple processes are faster. But the gains are at best efficiency. Most companies have only gone halfway; they haven't crossed the AI chasm… and I understand why. The full version is a far less comfortable place to sit, because it asks people to reimagine what their role would look like if they were removed from it, and asks the company to do the same to its departments, which is phenomenally difficult for any organisation to do to itself.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"And so, back to the middle. The big companies will continue to spend heavily on AI, mostly on a productivity agenda, and it will eventually show up on the bottom line, but I'm doubtful many of them will rewire the whole organisation from the ground up, because a fundamental burn-down of the kind Block is attempting is too bitter a pill for most. ","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"Opus Babies will do it from the very start, because they have nothing to burn. ","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"But for companies three or four years in, it is the most existential challenge of them all: refounding is not an optimisation but rather a necessity. A company in the middle has a finite runway to find and demonstrate its reason to exist, and it doesn't get to wait for the AI playbook to be written. ","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"Six months into ours, the progress has been phenomenal, we are far from finished, and we are far from the leaders in this space. But we are refounding ourselves in flight.","nodeType":"text"}],"nodeType":"paragraph"},{"data":{},"content":[{"data":{},"marks":[],"value":"Opus Babies were born this way. The rest of us get a second founding.","nodeType":"text"}],"nodeType":"paragraph"}],"nodeType":"document"}}}},"pageContext":{"slug":"an-agentic-refounding"}},"staticQueryHashes":["63159454"]}