How We Went All-In on AI Agents at Dawiso

Like most tech companies, we noticed the ChatGPT moment back in late 2022. Everyone started using chat AI for something — drafting emails, summarizing docs, brainstorming. Useful, but nobody's workflow changed. We were faster at some tasks. The tasks themselves stayed the same.

By mid-2025, we decided to get more intentional. We bought shared accounts — Claude, ChatGPT, Cursor. Made them available to every team. The message was simple: experiment, find what works, justify the ROI within two months.

No mandates. No AI strategy committee. One principle: adopt AI where it makes sense, don't force it where it doesn't.

The response was uneven, as expected. Some people jumped in. Others tried it once and went back to their existing tools. That was fine. We weren't looking for uniform adoption — we were looking for signal.

Through the second half of 2025, pockets of the team started going deeper. The development team introduced AI-assisted code review. Chat-based interaction became a natural part of the development process. We shipped a high-quality initial release of MCP integration for the Dawiso platform — giving AI agents a direct interface to our data governance layer.

None of this was revolutionary on its own. But the cumulative effect was clear: people who used AI daily were producing more, at higher quality, with less friction. The gap between AI adopters and everyone else was widening.

By the end of the year, we made a decision: AI agents would become part of every team. Not as tools on the side. As actual collaborators in the daily workflow — development, marketing, sales, customer success, back office.

In January 2026, we launched something we called AHA — AI Hyper Accelerator.

The name is half-serious, half-playful. We even made a logo and a song. But the purpose was dead serious: give people a structured space to break their mental blocks about AI.

We organized the first AHA in Prague. Two days, seven teams, voluntary participation. The brief was loose: pick a real problem, try to solve it with AI agents, share what you learned. We set expectations deliberately low on deliverables and high on learning.

What happened exceeded every expectation.

The projects varied wildly. A team tested five different AI tools for regenerating our website — work that, a month later, led us to rebuild the whole thing. Another team built a pipeline for generating conceptual data models from natural language descriptions, turning a multi-day design task into minutes. A third group implemented a complete user profile page using nothing but AI-generated code. Another experimented with AI-generated video production. Seven teams, seven different bets on where AI could move the needle.

But the real output was something else entirely: a shift in how people thought about their work. These aren't exceptional people doing exceptional things — these are regular team members discovering they can build production software.

That's the AHA moment. Not "AI is cool." It's:

I'll be honest: during the first AHA, I was too busy helping others to experience my own moment. So I carved out a weekend in February to do it properly.

In two days, I built a competitive intelligence platform from scratch. Serverless architecture, 12 database tables, automated web scraping of 15 competitors, complex scraping pipeline, AI-generated weekly reports and insights, Slack alerts. Running in production by Sunday evening.

The experience confirmed something I'd been suspecting: the economics of software development have shifted. The new bottleneck is specification, not headcount. How well your team can define what needs to be built — and how effectively they collaborate with AI agents to build it — determines output more than team size.

We ran AHA again in March, this time in Brno. More projects. More people. The pattern repeated — and the stories got sharper.

Our QA engineer built a bug detection toolkit. Prototype to working app in two and a half hours. Not a weekend. Not a sprint. An afternoon.

A colleague who had never opened a code editor built a personal notes app — attachments, tasks, Slack reminders, beautiful dark mode. Two days.

Another colleague built a music event tracker: Spotify integration, geographic concert feed, running in production. Her takeaway wasn't the app. It was that she could treat AI like a colleague: ask it how to ask it, give it feedback, and watch it learn.

A QA team member built a Slack bot that reads bug report threads and automatically creates structured Jira tickets with all fields filled in. It's been running in production since day one.

Across both events, common patterns emerged:

• Specification quality determines output quality. The teams that spent time on clear requirements got dramatically better results. AI amplifies the quality of your thinking.
• Templates and standards accelerate everything. Our internal app template gave every project authentication, design system, and deployment pipeline before a single line of business logic was written.
• Session discipline matters. The people who developed rituals — starting each session by loading context, ending by updating status — got consistently better results than those who just "chatted."
• Mistakes are transferable. When someone documented a problem and its solution, AI stopped making that mistake. Organizational learning became machine learning.

One thing we didn't plan for: the AI landscape kept shifting underneath us.

In January, we went all-in on Claude — both the conversational interface and Claude Code. The adoption was faster than expected. People who'd been skeptical became daily users within weeks.

By March, the GPT loyalists started resurfacing. New models, new capabilities. Our marketing team gravitated toward Gemini and Google AI Studio for creative work.

We stopped thinking about this as a vendor choice. The real question became: which AI for which task? Our developers live in Claude Code. Our marketers use Gemini for visual creative. Some team members keep GPT for specific workflows where it fits better.

Eight months in, the velocity keeps increasing. Every week, someone on the team does something that would have been impossible — or at least impractical — a quarter ago.

We're building things we aren't ready to talk about yet. Extensions to the Dawiso platform. New applications. New ways of working with our enterprise customers. The pace is unlike anything I've experienced in twenty years of enterprise IT.

What I can say: the biggest changes aren't behind us. The AHA moments keep coming, and each one raises the bar for what "normal" looks like.

If you're leading a technology company and you haven't given your team the space to experience this firsthand — structured time to break their workflow assumptions and rebuild them with AI — you're leaving extraordinary potential on the table.

We called it AHA because the name captures what matters. Not the technology. Not the tools. The moment when someone's eyes change and they say: "Oh. This is different."

That moment changes everything that comes after.