We are in the middle of the most significant platform shift in enterprise technology since the cloud. And unlike the cloud — which crept up over a decade — this one is moving in months.
I’m talking about AI agents.
🏁 The chatbot era is already over. Most people just don’t know it yet.
When ChatGPT launched, the conversation was about chatbots. “Can AI write my emails? Can it summarise this PDF?” Useful? Sure. Transformative? Not really. You were still the one steering.
Agents are a fundamentally different thing. An agent doesn’t wait for you to ask. It perceives its environment, reasons through a goal, takes actions — calling APIs, writing code, browsing, making decisions — and iterates until the job is done. Think less “smart assistant” and more “autonomous colleague.”
The difference between a chatbot and an agent is the difference between a calculator and a junior analyst. One answers; the other executes.
🔍 Here’s the real race — and it’s not who you think.
Everyone assumes the race is between OpenAI, Google, Anthropic, and a handful of well-funded labs. That is certainly a race. But the race I think about every day — the one that will determine which businesses survive and which ones get disrupted — is happening at the application layer.
It’s the race to build agents that actually work in specific, messy, real-world contexts.
• An agent that can autonomously manage procurement workflows for a mid-sized manufacturing company.
• An agent that screens and shortlists engineering candidates without needing a human to babysit it.
• An agent that monitors revenue operations, flags anomalies, and surfaces recommendations before your weekly pipeline review.
These are not research problems. These are engineering, design, and domain expertise problems. And they are being solved right now — often by smaller, faster, more focused teams than the hyperscalers.
At AT Dawn Technologies, this is exactly the space we’ve been building in. Not because it’s trendy, but because we believe deeply that the value in the AI era will accrue to whoever solves the last mile.
💡 Two business models will define the next five years.
I spend a lot of time thinking about the structural shape of this market. And I think it resolves into two dominant models:
1. Agent-as-a-Product — Companies that build specialised, high-reliability agents and sell access to other businesses. Think of it like SaaS but instead of selling software that requires a human to operate, you’re selling outcomes. Your agent runs the process; your customer just gets the result. The economic model is compelling, and the stickiness is unlike anything we’ve seen in software.
2. Agent-as-a-Moat — Companies that build agents not to sell them, but to run their own core business dramatically more efficiently than their competitors. If you can operate at 40% of the cost because your internal processes are agent-powered, you have a structural advantage that compounds. Your pricing power increases. Your margins improve. You can take on work that your less automated competitor simply cannot afford to.
The sharpest companies will eventually do both. But the first move is to get genuinely good at building agents.
⚠️ The hard part nobody talks about.
Here’s what’s missing from most of the AI agent discourse: reliability is hard.
Getting an agent to do something impressive in a demo is a weekend project. Getting an agent to do it correctly 98% of the time, in production, with real customer data, while handling edge cases gracefully — with the right fallbacks when it’s uncertain — that’s a different engineering problem entirely.
Memory matters. Context management matters. Knowing when to escalate to a human matters. Knowing when to stop matters.
The teams that will win this race aren’t the ones with the flashiest demos. They’re the ones who’ve done the boring, difficult work of making agents that enterprise customers will actually trust with their operations.
This is not a problem that gets solved by throwing more GPU hours at it. It requires people who understand the domain, the failure modes, and the user deeply. That’s a human problem as much as a technical one.
📢 Where does this leave you?
If you’re running a business and you’re waiting for this to “mature” before you start thinking seriously about agents — I’d push back on that instinct.
The companies starting to experiment now aren’t wasting time. They’re building intuition. They’re figuring out where agents break, which workflows are actually agent-ready, and how to integrate these systems into human teams without creating chaos. That knowledge is not something you can buy off the shelf in two years when everyone else decides to catch up.
The runway advantage is real.
And if you’re building in this space — if you’re one of the founders, engineers, or operators trying to make agents that actually work — I’d genuinely love to connect. This is a hard enough problem that the best thing we can all do is think out loud together.
The race is already on. The question isn’t whether to participate. It’s whether you’re building to lead it — or watching from the stands.