A Personal Agent in Every Pocket
Mustafa Suleyman's vision of AI as a personal companion isn't science fiction — it's the next platform shift, and founders are sleeping on it.
I was in a co-working space in Da Nang last year when a founder sitting next to me — building a legal-tech product — pulled up her AI assistant mid-conversation and asked it to summarise a 60-page contract she’d just received. It came back in 40 seconds with a structured risk breakdown. She didn’t look impressed. She looked bored. Like it was as ordinary as opening a calculator.
That moment stuck with me more than any conference keynote I’ve attended this year.
We are somewhere in the middle of a platform shift that most people are still treating like a productivity upgrade. It isn’t. It’s closer to the introduction of the smartphone — except the smartphone gave everyone a connected computer in their pocket, and what’s coming gives everyone a personal agent that knows them, remembers them, and acts on their behalf.
What Mustafa Suleyman Actually Said
Mustafa Suleyman — co-founder of DeepMind, then founder of Inflection AI where he built Pi (Personal Intelligence), and now CEO of Microsoft AI — has been the clearest voice about where this is heading. In his 2023 book The Coming Wave, he wrote:
“Whatever your job, you will be able to consult an on-demand expert… All of the world’s knowledge, best practices, precedent, and computational power will be available, tailored to you, to your specific needs and circumstances, instantaneously and effortlessly. It is a leap in cognitive potential at least as great as the introduction of the internet.”
That’s not the language of someone describing a better search engine. That’s the language of someone describing a fundamental restructuring of how humans access knowledge and take action.
When he joined Microsoft in March 2024 to lead its consumer AI division, he framed his mission as building “an AI companion for everyone.” The Copilot upgrades he’s been overseeing — persistent memory, multi-session context, personalisation — are the early scaffolding of that vision. He also noted that every device you own — tablets, screens, cars, fridges — is becoming a “conversational endpoint.”
Suleyman is not describing a chatbot. He’s describing a persistent, personal layer of intelligence that rides across every surface you touch.
This Is a Platform Shift, Not a Feature
I’ve built products on every major platform wave — web, mobile, cloud. Each wave had the same shape: a new runtime appears, most people treat it as a new channel for old business logic, a small group of founders treats it as a new substrate, and the substrate builders win.
In the web wave, the losers were companies that put brochures online. The winners built native web businesses — Amazon, Google, eBay.
In mobile, the losers were companies that shrunk their websites. The winners understood that a phone is always on, always located, always personal — and they built Instagram, Uber, WhatsApp.
In the agent wave, the losers will be companies that add a chat interface to their existing SaaS. The winners will understand that an agent is always available, knows your context, maintains continuity, and can take action across systems.
The key difference from every prior platform shift: agents are stateful and proactive. They remember. They initiate. A smartphone app waits for you to open it. An agent running on your behalf doesn’t wait for anything.
What “Personal” Actually Means Here
The word “personal” is doing a lot of work in Suleyman’s framing, and it’s worth unpacking.
Pi — the chatbot Inflection built before the Microsoft deal — was deliberately designed around emotional continuity and memory. It remembered past conversations. It got to know users over time. Suleyman described its long-term goal as being a “digital Chief of Staff.”
I’ve had a digital Chief of Staff at various points in my career — a real human one. What that person actually does is anticipate, pre-filter, contextualise, and smooth. They know your preferences before you articulate them. They know which investor email is worth your attention at 11pm and which one can wait until Monday. They know when you’re overcommitted and they push back on your behalf.
Getting that kind of intelligence scaled to every founder, every operator, every person — not just those who can afford a team around them — is genuinely radical. Not in a TED-talk way. In a structural redistribution of leverage kind of way.
When I talk to early-stage founders in Bangalore or Kochi, the ones who are moving fastest right now aren’t the ones with the biggest teams. They’re the ones who’ve figured out how to use AI context management effectively — keeping an agent loaded with their product context, their market context, their personal decision patterns. It’s an approximation of what Suleyman is describing, and even the rough version is compressing what used to take months into days.
The Technical Reality in 2026
To be clear about where we actually are versus where we’re heading: agents today are still brittle at the edges. Long-context windows have improved dramatically — we went from 8K to 1M+ tokens across frontier models. Memory is being solved architecturally with retrieval-augmented approaches and episodic stores. Tool use and multi-step planning have gotten much more reliable.
But true persistent personal agents — ones that run continuously, maintain genuine user context across months and years, coordinate multi-system actions without human supervision, and make trust-worthy decisions in ambiguous situations — that’s still being assembled. The infrastructure is arriving in pieces.
What this means for builders is that the window to establish position is now, not when the full system is polished. The same was true for apps in 2008 — the App Store was buggy, the SDK was limited, battery life was terrible — but the people who understood the platform early and built real utility captured the high ground.
Where the Actual Opportunities Are
I see three categories of real opportunity in the personal agent shift, and I’m actively thinking about all three as I build Zero (zero.university):
Identity and continuity infrastructure. Agents need somewhere to store and retrieve what they know about you. Not a cloud folder — a structured, permissioned, personal knowledge graph. Whoever becomes the canonical store of personal AI context is sitting on something enormous.
Agent-native applications for high-stakes domains. Health, finance, legal, education. These are areas where the gap between “person who can afford expert advice” and “person who can’t” is massive. A personal agent that can navigate your insurance, track your health trends, explain your tax position, or tutor your kid at their exact level — that’s not a feature, that’s a new category. I’ve been thinking hard about this for Zero specifically.
Trust and transparency layers. As agents start acting on your behalf — booking things, paying for things, communicating on your behalf — the question of what they’re doing and why becomes critical. Verifiable audit trails, user control interfaces, agent accountability — this is unglamorous infrastructure that will be table stakes within three years.
The Founder I Keep Thinking About
I had a conversation in Dubai last winter with a founder building an agent specifically for real estate agents in the Gulf — not for property buyers, for the agents themselves. The agents she was targeting have massive context loads: client histories, property inventories, regulatory nuance across Emirates, relationship dynamics with developers. Her insight was that the agent’s value wasn’t in answering questions, it was in knowing what questions to anticipate.
Three months after launch she had 200 paying users. Not because the AI was magic — the underlying model was a commercial API — but because she’d done the hard work of understanding what a real estate agent in Dubai actually needs to hold in their head across a working week. The domain specificity was the moat.
That’s the pattern. Suleyman’s vision becomes real one vertical at a time, built by people who understand both the technology and the lived reality of a particular profession or context.
The founders who will win the personal agent era are not the ones who think most abstractly about AI. They’re the ones who understand one human workflow deeply enough to know exactly what an agent that truly knows someone in that workflow would need to do.
That’s always been the game. The platform just changed.
