From Chat to
Orchestration in
635 Days
I co-founded Vu Agency on 7 June 2024. In the 635 days since, AI went from a thing you type questions into, to a thing that does your job while you watch.
I co-founded Vu Agency on 7 June 2024. I know the exact day count because someone put "635 days" in the title of this post and I didn't stop them.
Here is what happened in those 635 days: AI went from a thing you type questions into, to a thing that does your job while you watch. And somewhere in between, Alex and I started a company, won an award we didn't expect, built three products, and redesigned our own website by letting strangers on the internet vote on it.
I'll get to all of it.
Some background
Alex and I have known each other for years. I was the first employee at his previous company. We worked together, went separate ways, kept talking. By early 2024 we were both using AI every day and noticing the same thing: work we used to do in a week was getting done in a day. Not small tasks. Proper deliverables.
The obvious move was to help other businesses see the same thing. Most companies we spoke to were still treating AI like a chatbot you poke when you're bored. The gap between "what AI does" and "what businesses think AI does" was enormous. We incorporated Vu AI Ltd on 7 June 2024 and started filling the gap.
Nine months later, we won Breakthrough Business of the Year at the Solihull Chamber of Commerce Awards. We did not see that coming. But it was a good sign the work was landing.
The website thing
A few weeks ago, we redesigned this website. We used 102.ai, our own AI-powered web design tool, to generate three completely different designs. Then I put all three out to our community and asked people to vote.
The design you are looking at right now is the one they picked.
Here are the other two:
The votes were close. All three were strong. But the community chose this one, and we went with it.
This is a small example of how we think AI should work. The AI generates the options. A human makes the decision. Nobody got replaced. The process got faster.
102.ai and Raq.com
Two products came out of client work we were doing.
102.ai is our AI web design tool. You describe what you need. Your site is designed (or redesigned) and live in 102 minutes. We launched it roughly a month before Raq.com.
Raq.com is broader. It is a single platform with AI chat, e-signatures, document design, forms, image generation, rewriting tools, and more. Pay-as-you-go. No subscriptions stacking up. We have been shipping new tools on it every week since launch, and it is now almost a month old.
Both started as tools we built to solve our own problems. Then we realised they should exist as products.
What we have been building for clients
Between the products, we have been building AI-powered platforms for businesses across completely different industries. A KYC compliance system for a finance broker. An AI carbon literacy assessor for an education company. A full operations and workshop management platform for a fitted bedroom business. An AI booking assistant for a Birmingham events venue.
Different industries, same pattern. AI does the repetitive bit. Humans stay in control of the decisions.
That is the thread running through everything we do, and it is the same thread running through everything that happened in AI over the past 635 days.
Now here is what changed everywhere else
All of the above was our little corner. Meanwhile, the entire AI landscape shifted underneath everyone's feet. Five things stood out.
Five things that changed while we were building
All of the above was our little corner. Meanwhile, the entire AI landscape shifted underneath everyone's feet.
SCROLL TO EXPLOREReasoning models showed up
In September 2024, OpenAI released o1. The idea was simple: instead of answering instantly, the model spends more time thinking. Instead of predicting the next word, it checks its own work.
Anthropic followed with Claude 3.7 Sonnet. Google launched Gemini 2.5 as a "thinking model." By August 2025, GPT-5 turned the whole thing into a unified routed system. The competition moved from "which model sounds the most fluent" to "which model solves the hardest problems." Different game entirely.
Agents started doing the work
Anthropic launched public computer use in October 2024. OpenAI released Operator in January 2025. A system built to click, type, and browse on your behalf. Google framed Gemini 2.0 as a model for the "agentic era."
Think of it like driving on a motorway. You hit a straight stretch, you tap autopilot. The car handles the boring bit. When the road gets interesting, you take over.
AI stopped answering questions and started doing work. Multi-step work. Research, draft, build, check, iterate, ship. Your role does not disappear. Your role changes. You stop doing the work. You start directing it.
AI started writing real code
Claude 3.5 Sonnet hit 49% on SWE-bench Verified in October 2024. GitHub rolled out Copilot agent mode. OpenAI launched Codex as a cloud software engineering agent. AI coding tools became the fastest-growing category in developer tooling.
Developers reported massive productivity gains. And here is the part nobody predicted: they worked more, not less. When you ship in hours what used to take weeks, you do not down tools and go home early. You take on the project you would have turned down. You build the thing you filed under "someday." More capacity does not reduce work. More capacity reveals how much work was always waiting.
Open models changed the economics
Meta released Llama 3.1 405B in July 2024. Then DeepSeek-R1 arrived in January 2025, claiming performance comparable to OpenAI o1 while open-sourcing everything.
Nvidia lost roughly $589 billion in market value in a single day. The moment a lot of buyers stopped assuming frontier AI would stay closed, scarce, and expensive.
Search became AI. Regulation landed.
Google's AI Overviews scaled to more than 1.5 billion users by May 2025. Publishers filed antitrust complaints, arguing AI summaries sit above links and hurt traffic. For agencies and businesses, this is one of the biggest commercial changes of the period.
Meanwhile, the EU AI Act entered into force on 1 August 2024. Prohibited-practice rules started applying in February 2025. AI governance moved from "we should think about this at some point" to live compliance work.
What this means if you run a business
In 635 days, AI went from a tool you experiment with to infrastructure you need a strategy for. Three things now matter a lot.
Three things you need to know
SCROLL TO EXPLOREModel choice is a procurement decision
Dozens of capable models exist. Each with different strengths, pricing, and trade-offs. Choosing between a fast model for customer-facing chat and a reasoning model for complex analysis is not a technical curiosity. It is a CFO-level question.
Get it wrong and you are overpaying for capability you do not need, or under-investing in the tasks where AI matters most.
Workflows are moving from chat to orchestration
The prompt-and-response pattern was the start. AI agents now run multi-step workflows: research, draft, validate, iterate, deliver. A salesperson orchestrates agents to research accounts, write personalised outreach, and prep meeting notes. A marketer orchestrates agents to generate campaign variants and draft landing pages in real time. A finance lead orchestrates agents to reconcile data, model scenarios, and flag anomalies before anyone asks.
The businesses seeing real returns from AI are not the ones with the best chatbots. They are the ones with the best orchestrated workflows.
Governance is a competitive advantage
The EU AI Act is in force. The UK is accelerating its own regulatory framework. The businesses who get governance right will build trust, move faster, and outpace competitors still scrambling.
Good governance is not a brake. Good governance is a licence to deploy at scale.
A prediction
I will put a stake in the ground here: within two years, most knowledge workers will orchestrate AI agents as a normal part of their day. The same way you use email now without thinking about it. That is where this is going.
Your marketing person will not write the first draft. They will brief an agent, review the output, and direct revisions. Your finance team will not manually reconcile spreadsheets. They will set agents to run the reconciliation overnight and review the exceptions in the morning. Your operations lead will not chase updates from five departments. They will check a dashboard that agents keep current.
The people who learn to direct agents well will be disproportionately valuable. The ones who wait to be told how will be catching up for years.
This is not a guess. The tools exist today. The cost is already low enough. The only thing missing is adoption, and adoption is accelerating.
Where we are now
We built Vu in the middle of all this. Not to chase trends, but to help businesses make sense of them. Every tool we build, every report we generate, every recommendation we make is grounded in what works right now. Not what might work someday.
We recently got shortlisted for the Digital Revolutionary Award at the Greater Birmingham Chambers of Commerce Awards 2026. Between this and the Breakthrough Business win last year, it feels like the work is speaking for itself. But we are not slowing down.
102.ai is live. Raq.com is live and growing weekly. And the free AI business analysis tool is still the fastest way to see what AI means for your specific operations.
If you feel behind, most people do. The barrier is not technical. People feel overwhelmed, so they do nothing. And doing nothing is the one move guaranteed to put you on the wrong side of this.
Enter your website below. In minutes you get a full report on the opportunities specific to your industry and operations. No cost. No commitment.
Phil Webb is CEO and Co-Founder of Vu Agency, 102.ai, and Raq.com. Based in Birmingham. Partial to Leicester Tigers, strong cortados, and building things nobody asked for but everyone ends up using.