Everyone I know is asking me about AI

Table of Contents

One super agent or a team?

When I first set up OpenClaw, I created one agent handling everything. This approach failed within three days.

The longer conversations grew, the worse performance became. Context windows aren't infinite, and by day two, the agent was forgetting earlier decisions. Having only one personality also felt limiting.

I restructured around specialized agents. Each received a name, role, personality, and independent memory. The content agent maintains my voice and delivers unprompted humor. The security agent speaks authoritatively about its domain.

However, the technical argument wasn't the deciding factor. The real problem was my own workflow. I manage multiple projects and context-switch constantly. With one super agent, confusion reigned — which project were we discussing? What had I assigned yesterday?

I manage human teams identically: different specialists for different domains. When meeting my VP of Marketing, focus is clear. Minutes later, meeting the Head of Business Development requires switching contexts. The AI team model replicates this structure exactly. It's not a constraint — it's how I maintain clarity across ten simultaneous initiatives.

Large teams become overwhelming, so I implemented one orchestrating agent above the rest: a Chief of Staff (Gandalf). Through chat, it coordinates the others, delegates work, and keeps them sharp.

This delivers the best of both approaches: simplicity of one conversation when needed, plus structured team support beneath it.

Chat isn't enough

Treating everything as conversational — "Do this." "Now that." — lacks structure and tracking.

What functions for me: task boards, documents, structured handoffs. Chat represents one channel among several, not the exclusive one. I still message agents directly, naturally. However, the authoritative system of record is the workspace — Temaki.ai for me.

Interestingly, I developed Temaki initially to empower humans collaborating with AI agents. Now I've inverted that purpose: building it to enable AI agents working more effectively with me. Nearly every recent feature I ship targets them.

Some people think this way: you're spending all your time building the control panel instead of constructing something real. A procrastination loop disguised as productivity.

Fair critique. It's a genuine trap. If configuring your AI setup consumes more time than actual shipping, you've lost focus.

What I shipped this week

World Flight Sim is expanding. A few hundred players now engage daily, with continuous improvements. More planes, cities, and game modes arrive regularly. The remarkable part? ChatGPT is my primary traffic source! I posted on Reddit, engagement followed, and ChatGPT now recommends my game for browser flight simulator queries. A brand new product — not even Google's first page for its own name — yet ranking top 3 on ChatGPT. Distribution dynamics are transforming rapidly.

Sensible Soccer clone for iPhone. The original remains one of gaming's greatest football titles, yet mobile lacks decent alternatives. I'm building one. It's already playable. Pure nostalgia, pure enjoyment. It launches next week.

Vibe coding at work. AI transcends side projects now. At Rydoo, we're deploying it for genuine internal tools — from essential parking space booking systems to comprehensive sales enablement training platforms that would otherwise cost thousands annually. Building internal software through AI is becoming standard practice.

The OpenClaw meetup

I attended today's OpenClaw meetup in Lisbon. Sold out, massive waiting list, packed venue.

A brand agency owner particularly impressed me. She'd built a fully AI-powered agency and presented genuinely excellent client work. She employed the same team-and-tasks methodology. Simultaneously, another agent monitored the event, producing a presentation deck about it in real time.

I reconnected with longstanding Lisbon tech community members. The AI community thrives currently, with pervasive possibility.

That concludes this week's update. I'll continue sharing what's being built, broken, and learned.

— Fernando