Fourteen years ago I wrote a blog post called "We are five years old. Here is what we learnt." It ended with a promise: "Here's to the next five."
That was in 2012, and I guess I was planning to write an update five years later in 2017, when we turned ten. Well, this is that post, a mere nine years late. What can I say? We’re better at making things than marking occasions. But the point is, we were nineteen yesterday, and here is what we learnt.
So, let's start by going back in time. Again.
It's September 2007. The global financial crisis is just a few weeks from becoming apparent. Four naive idiots are un-boxing new Apple equipment behind a wall of glass in BBH London's marble atrium, joking nervously about keeping the boxes so they can sell the machines when the company folds.
They have a set of hypotheses that sound, in 2007, genuinely eccentric. That it's possible to make agile work for strategy and design, not just software development. That the future belongs to products and services, not campaigns. That open source and cloud computing mean a company can be as light as air. That small teams making real things will beat large teams making documents. That the most exciting digital work isn't coming from agencies at all, but from start-ups, because start-ups make stuff people actually use. Stuff they love.
They put a provocation slide near the front of the creds deck and used it to test potential clients. If a client flinched at the slide, they wouldn't work with them. Some of the early decks had a Star Wars theme. No apologies.
I wrote in 2012 that "we were right to focus on products and services, but a little early." As usual, I had no idea. We weren't a little early. We were nineteen years early.
What we got wrong, quickly
The 2012 post owned up to the hypotheses that failed, so this one should too.
We lasted exactly twelve months trying to work without our own developers. The plan was to be technology-agnostic and partner for engineering. It collapsed for a reason I described in 2012: "You can't make stuff at the same speed you can dream it up and test it if your devs are sitting somewhere else."
We said projects, never retainers, then discovered that when the work goes well clients want to buy a team, and softened. Really like retainers these days. We named a methodology, then another, then a few more, and eventually admitted we'd made "all the rookie errors there are to make, not least the really stupid one that involves repeatedly fetishising one methodology after another."
And more than once we drifted. All over the place. Towards being an agency, towards big transformation, towards growth by headcount. Every time, we came back to the same small, stubborn shape: one senior team, multiple hats, no silos, making the real thing, letting evidence beat opinion even when the opinion was ours. We wrote in 2015 that "the business and the company are prototypes for defining a better model." Nineteen years in, that's never been more true.
The thing that never changed
Underneath every era of this company sits one problem, and it's the one in that 2012 sentence: the gap between dreaming something up and making it real.
Everything we ever built has been an attempt to close this dreaded gap. Bringing developers in-house closed some of it. Prototyping from week one closed more: by 2012 we could say that prototyping "means we can finally make stuff at the same speed we can dream it up and validate it," and a journalist writing about our ITV News work noticed that we "had not just prototyped tools, but prototyped workflow as well." Fused teams, dedicated to one problem instead of working across many. Short phases. Embedding for real-time decision-making instead of sending documents into the machine and waiting for answers. All of it was the same fight against ‘the dreaded gap’.
We closed the gap as far as possible, but designing something and building it remained two acts with a handover, and even when you’re doing both bits, the handover is where good ideas lose momentum.
Then the gap started to dissolve
I want to be honest about this, because I can't help it: we did not see AI coming. Not like this. Nothing we wrote before 2024 mentions it. We are very cynical about tech-hype. It’s a coping strategy if you grow up in a revolving hype cycle. This is not a prophecy post.
But, here's a truth stranger than prophecy. For nineteen years we focused all that energy on the gap problem, and then a new technology arrived that just dissolved it. Designing and making have finally become the same act, done by the same small team, at the speed of thought. The thing we could only approximate through method and sweat is now just how the work happens.
AI is the best thing that ever happened to a company like ours. Not because it changes what we believe, but because it stops us having to compromise. The reason we couldn't stay small, flat and senior at enterprise scale was that making at that scale needs bodies. Now, it doesn’t. The 2007 hypotheses weren't wrong. They were waiting.
Anyone can generate an output now. Knowing whether the output is any good, or worth building at all, is a different problem. And it's the most fundamental problem to exercise us since we started, back before the iPhone even had an App Store. (Yes, we are *that* old.)
So we wrote the judgement down
Which brings me to the tease. This year we did something we probably should have done a decade ago: we wrote down what we actually do. Not as another sacred methodology, but as a set of working instructions that AI agents run. We call it our Agentic Playbook, and it's how nineteen years of practice becomes something an agent can execute rather than a bunch of documents that people skim.
There is a lot more to say about it soon. But for now, four facts serve as a taster of what's inside:
- A full agentic SDLC. Thirty-two skills across nine stages, running greenfield from concept to production. Opinionated about what makes software good, undogmatic about how you get there, and built to adapt to a client's stack, process and team rather than demand they adopt ours.
- Studio skills for everyone. A second suite, installed across the whole organisation for daily work, because the leverage shouldn't stop at the delivery team.
- A knowledge MCP. Ten years of the studio's work, indexed and retrievable through one server that any tool can talk to. The agents don't just have our instructions; they can consult our experience.
- A learning loop. Every skill is a snapshot of what we know now. What each project teaches us goes back into the playbook, so the skills grow and mature alongside the people who use them.
In 2012 I wrote that the hardest lesson we'd learnt was how to learn, and that we saw rapid learning as a core part of the offer. The playbook is that lesson made executable. We no longer have a gang of hairy-arsed devs sitting here, or anywhere. Today, they don’t even have arses.
Still here, still awkward b*stards
Nineteen years. Still independent. Still making. Still testing ideas against reality instead of arguing about them in the abstract. Still, I'd like to think, the kind of company that would put the flinch slide at the front of the deck.
In 2012 I signed off by saying the post was in beta and my colleagues would chip in with lessons I'd missed. Both things are still true.
Here's to the next nineteen. We'll write it up in 2045, if I’m still alive.
