// on AI
I believe we're in the middle of the biggest shift in how software works since the smartphone. Clicking around is dying. The future belongs to agents that understand intent and act on it.
// what I believe
The next generation of software won't have menus, tabs, or dashboards. You'll just tell it what you need. Need data? Ask. Want to trigger a workflow? Say so. Send an invoice, freeze a card, spin up an environment — one prompt, done. Companies like Revolut are already proving this: AIR doesn't sit beside the banking app — it replaces the interface entirely.
Traditional automation handles the predictable. Agents handle the unexpected. An email agent doesn't just sort messages — it reads context, drafts replies, and escalates when uncertain. A voice agent doesn't just route calls — it resolves issues, pulls data, and acts on behalf of the customer. The difference is autonomy with purpose.
The biggest unlock isn't standalone AI tools — it's embedding intelligence into existing business processes. When an agent sits inside an orchestrated workflow, it gets guardrails, audit trails, and human oversight for free. That's how you get AI into production at enterprise scale, not as a demo.
// how I work with AI
Start with a clear specification, let AI generate the implementation. Review, iterate, ship. This isn't about replacing developers — it's about amplifying their intent. I use Claude and Copilot daily to turn specs into working code at 3-5x the velocity.
Every agent I build follows the same principle: clear boundaries, defined tools, structured output. Whether it's an email responder, an SMS handler, or a voice agent on Retell or VAPI — the architecture is consistent. The agent knows what it can do, what it can't, and when to ask a human.
Don't bolt AI onto existing interfaces. Design for conversation from the start. This means rethinking information architecture, user flows, and what 'interaction' even means. The best AI products feel like talking to a competent colleague, not clicking through a wizard.
Demos are easy. Production is hard. I focus on error handling, fallback strategies, cost management, and observability from the first line of code. Using AWS Bedrock, OpenAI, and Claude — choosing the right model for the right task, not the most expensive one for everything.
We're not just adding AI to existing products. We're rethinking what products are. An ERP system shouldn't be 200 screens — it should be one conversation. A customer service platform shouldn't be a ticket queue — it should be an agent that resolves issues before anyone notices them.
I'm building toward this future — both in my work and in what comes next. The companies that understand this shift early will define the next decade of software.
Want to talk about AI, agents, or building the future of software?