Pushing the Boundaries of Building Innovative Tools: Key Learnings from Toqan & Otomoto
Pushing the Boundaries of Building Innovative Tools
Key Learnings from Toqan & Otomoto
Building AI-powered products at scale teaches you things no framework or blog post can. These are 8 lessons we've learned the hard way while building Toqan and the Otomoto Dealers Agent.
1. Value Comes First — Technology Is a Means, Not the Goal
Every new technology introduces both advantages and constraints. The winning approach is to apply AI where it is structurally superior.
In Otomoto Dealers Agent, AI was most effective at rapidly assessing large volumes of diverse data and translating them into simple, actionable insights for dealers.
Don't start with the technology. Start with where it creates asymmetric value.
2. Visual Interfaces Outperform Text by Default
Moving from a text-heavy interface to an action- and button-first design resulted in a 4x increase in engagement.
Visual cues reduce cognitive load, accelerate decision-making, and allow users to extract value faster. If your agent is outputting walls of text, you're leaving engagement on the table.
3. Embedding Innovation into Existing Platforms Is Harder Than Building Greenfield
Mature platforms are already highly optimized. Introducing new solutions requires carefully balancing intrusion into existing workflows with the delivery of exponential value.
Incremental additions rarely justify the real estate they consume. If you're going to embed AI into an existing product, make sure it earns its place.
4. The Pareto Principle Strongly Applies to AI Agents
Roughly 20% of agents generate the majority of value, while the remaining 80% provide incremental or supportive benefits.
Identifying and doubling down on these high-impact agents is critical for focus and ROI. Not all agents are created equal — know which ones matter.
5. High-Impact Agents Demand Engineering Rigor
The most valuable agents typically require deep engineering involvement. Secure integrations, robust authentication, and reliable data access are non-trivial and cannot be abstracted away without compromising impact.
If your highest-value agent is held together with duct tape, you have a problem.
6. Long-Context Handling Is a Core Scaling Constraint
At scale, long-context management becomes decisive — not only due to cost, but because it directly determines the complexity and depth of tasks the system can perform.
Architectural decisions here compound over time. Get this wrong early and you'll hit a ceiling you can't easily raise later.
7. Users Who Hack Your Product Are Your Best Signal
Power users who build workarounds or stretch the product beyond its intended use reveal unmet needs and latent demand.
Actively engage them, understand their mental models, and turn their hacks into first-class capabilities. These users are telling you what to build next.
8. Innovation Is About Leverage, Not Coverage
Success comes from concentrating effort on areas with asymmetric returns — where technology, user pain, and workflow leverage intersect.
Broad feature coverage without clear leverage dilutes both impact and speed. Focus beats breadth, every time.
The Bottom Line
Building innovative AI tools isn't about having the best models or the most features. It's about understanding where technology creates real leverage, designing for how humans actually work, and having the discipline to focus on what matters.
These lessons came from building Toqan and Otomoto — and they'll shape how we build everything next.