AI Founder Charlie Cowley

I last met Charlie Cowley when he was building Impala, the hotel API platform he co-founded in 2016 that went on to raise around $35 million, scale to 120 people, and work with some of the biggest names in global travel. That company didn't end the way anyone planned. The scars, by Cowley's own admission, run deep. Letting people go. Tough shareholder conversations. The kind of darkness founders rarely talk about publicly. So, when I heard he was building again - same co-founder, overlapping team, new sector - I wanted to understand what pulls someone back into the ring.

This morning's venue wasn't chosen by Cowley but by Sophie - his Dream-powered personal assistant. She sends us to Catalyst Café on Gray's Inn Road, a Scandi-style specialty coffee spot in Holborn where they roast their own beans in the basement. Sophie clearly has good taste, and - fortunately for me this morning - a light touch on breakfast: we've opted for two pain au chocolats between us rather than a larger spread…. excellent for the budget. Cowley arrives with his arm in a sling, collarbone recently broken playing football and now surgically repaired, organising his week almost entirely by voice note through the very product he's building. It's a fitting, if slightly theatrical, proof of concept.

Dream is an AI assistant platform for busy professionals & small businesses, recently backed by SpeedInvest in a seed round. Not another chatbot. Not another dashboard. The pitch is action, not conversation - assistants that send emails, book restaurants, update calendars, coordinate travel, and when the AI reaches its limits, spin up real humans behind the scenes to complete the task.

It's the action, the task, that people want value from. Not just another conversation with GPT.

The starting block is hyper-specific: AI Personal Assistant (for busy professionals) and AI Business Assistant (for small businesses). Dream's first consumer-facing product will be the personal assistant for London professionals - booking tables, sourcing flowers, coordinating dry cleaning, handling the kind of concierge tasks that previously required either a £20,000-a-year PA or a lot of personal admin. The idea is to prove the model in a single city, with deep local context, before expanding to Paris and beyond. "It's the action, the task, that people want value from," Cowley says. "Not just another conversation with GPT." Together, they form what he sees as the operational bedrock for any lean business or busy professional.

I push him on defensibility. The AI agent space is crowded, noisy, and moving at a pace that makes last month's innovation feel dated. Cowley is clear-eyed about the challenge. Dream has spent nine months building a platform layer that combines deep personal context - who you are, what your business does, how you like things done - with tooling that enables the assistant to actually execute. When the AI can't complete a task, a human-in-the-loop steps in, managed by the AI itself. It's a hybrid model: part software, part operations, designed to deliver outcomes rather than suggestions.

Once you build up that level of trust and it performs, you're off to the races.

There's something quietly radical about giving an AI assistant a name. Cowley admits the team was initially nervous about it - the psychological implications of attachment, reliance, even protectiveness. But what they've found is that a nickname, chosen personally by the user, accelerates trust dramatically. "Once you build up that level of trust and it performs, you're off to the races," he says. It's a small design choice with outsized consequences.

The conversation turns to Impala, and what building a second company with the same core team really means. Cowley and his co-founder Ben Stephenson started Dream together; Philip and Chris, early hires at Impala, came across too. They know how each other work under pressure. They've made the mistakes already. That institutional memory, Cowley argues, compounds - in speed, in decision-making, in knowing what not to do. "We have the battle scars. We have the learnings. Let's apply that to this one."

But he's honest about the counterweight. A team of thirty-somethings with families and different life stages doesn't carry the same raw energy as a group of twenty-four-year-olds. The 996 culture - nine to nine, six days a week - doesn't map onto a team with children and partners and a different relationship with time. Cowley doesn't romanticise the grind. "Some of the best work gets done going for a walk and just thinking through how you're approaching a problem," he says. The hours still have to be long, but they must be considered. At Impala, he prioritised work over everything - no proper holiday for four or five years, the gradual breakdown of boundaries. He's doing it differently this time.

If you're trying to build that next big thing, you need your own level of arrogance about what you're building. Otherwise, you end up regressing to the mean.

One of the sharpest lessons from Impala, he says, was favouring long-term vision over short-term execution. It sounds counterintuitive - isn't long-term thinking supposed to be the hallmark of a serious founder? Perhaps… but Cowley's point is more specific. In a world moving this fast, particularly in AI, getting lost in a grand five-year roadmap can blind you to what needs to happen this month. Revenue. Product. Optimised, tangible progress. Eyes on the road.

On building in the UK, Cowley is measured. The barriers to starting a company are low - SEIS and EIS remain among the best tax incentives for early-stage investment anywhere in the world. His wife started a handbag business alongside her job during Covid, and the ease of doing that shouldn't be taken for granted. But he's frustrated by the gap between starting and scaling. The UK incubates well but struggles to provide the framework - capital, consistency, conviction - that founders need at the next stage.

What irritates him most is the lack of clarity from government. Not the tax rates themselves, but the uncertainty. Even the rumour of a 50% capital gains hike in the last Budget was enough to rattle founders weighing up whether to stay. "You just need someone to come in, make the decision, follow through, and be clear," he says. "That's going to be the plan for the next five years. Rather than - I have no idea." He draws a light-hearted comparison to Trump: whatever you think of the politics, there's a directness and follow-through that gives people confidence to at least plan. The UK, by contrast, feels governed by politicians more afraid of losing their jobs than building long-term prosperity.

The deeper thread running through the conversation is one of cognitive outsourcing - what happens when we hand more and more of our thinking, our admin, our decision-making to AI. Cowley is building tools that accelerate exactly that, and yet he's thoughtful about the limits. Human connection, he says, is the obvious thing not to outsource. At Impala, the team worked fully remotely during Covid, and something essential was lost. You need to be able to turn around and have a beer at the end of a hard day. You need the unstructured, unoptimised moments.

You always want the end to sit within your own grasp.

His view is that AI should handle eighty percent of the heavy lifting - freeing people to spend more time on relationships, creativity, and the things that actually matter to them. But intent is key. "You always want the end to sit within your own grasp," he says. The blur between AI that amplifies human capability and AI that starts making decisions for you is where things get uncomfortable. Dream is building in that tension deliberately: proactive assistants that act on your behalf, but with you retaining control over what matters.

I ask about the speed of change, and whether even a founder building in AI can keep up. He laughs. "You can't keep up with absolutely everything." New models, new platforms, new competitors - the landscape shifts weekly. A company called Rent-a-Human recently launched a platform where AI agents can hire actual people to complete physical tasks. It's both thrilling and unsettling. Cowley's own engineering team uses AI to write code that would previously have required a team three times the size. The leverage is real. But so is the vertigo.

As we wrap up, I ask for the maxim he lives by. He references "strong opinions, loosely held" - the framework coined by Silicon Valley forecaster Paul Saffo in the 1980s and since adopted as gospel across the tech world. The idea is simple: form a conviction quickly based on available evidence, then actively try to disprove it. Stay flexible. Update your view. Cowley firmly rejects it. "I don't agree with the loosely held part," he says. His version is strong opinions, strongly held. In a world full of noise, market shifts, and competitors building in every direction, a founder's job is to maintain conviction in their vision - not to be swayed by whatever everyone else is doing. "If you're trying to build that next big thing, you need your own level of arrogance about what you're building," he says. "Otherwise, you end up regressing to the mean."

It's a philosophy that carries risk - and he knows it. Sometimes the data will prove you wrong, and that's fine. But ninety percent of the time, it's conviction that sustains the iteration, the knock-downs, the slow route to building something that generates real value. With that, we're done. Cowley heads back out into a grey March morning in Holborn - arm in a sling, week managed by an AI agent, building his second company with the same team and a harder-won clarity about what matters. Dream is in private-beta and moving fast. If the product works as advertised, the irony is rich: a founder whose own assistant booked the café, coordinated the calendar, and handled the logistics - leaving him free to sit across the table and talk, human to human, about why that still matters most.

Visit Dream