Why Todoist says less AI can deliver more

Doist CTO Gonçalo Silva emphasizes that the pace of AI change has made it difficult to commit to one roadmap, model provider, or way of working. Rather than predicting which direction will win, Doist, the company behind Todoist and Twist, established three broad principles over a year ago: purposeful AI features, private and secure AI, and adaptability as models change.

Doist has found that just because a feature is easy to build with AI doesn't mean it's worth shipping. The company is currently testing Automations, a service that turns natural-language requests into executable workflows. While AI interprets user intent, ordinary code executes the steps. However, Silva warns that AI is not predictable and consistent by design, which poses issues for consistent automation.

Keeping AI models out of the execution path is advantageous as it's cheaper and faster to run code on a CPU compared to invoking expensive models repeatedly. This approach also reduces costs and speeds up execution.

Doist's broader product philosophy is "subtraction over addition," focusing on eliminating features that haven't proven their worth. Todoist, for instance, recently retired its Goals beta due to poor usage data and experience.

The company is cautious about adopting AI features, ensuring a quality bar is maintained. They test AI changes by checking behavior against multiple models and languages. This approach has also helped control inference costs, which became an issue when an early beta of Automations resulted in a significant bill during its initial rollout. Ultimately, Doist acknowledges the need for expert judgment in evaluating AI-generated features, even with the autonomy provided by AI tools.