Dominic Williams outlines next phase for Caffeine after user feedback

A candid post from Konecta DAO founder and product manager @varon1980 has triggered a broader conversation about usability, expectations and communication around Caffeine, the AI-powered development tool built on the Internet Computer.

Writing on X, the long-time ICP investor described his comments as constructive rather than critical. He said he has supported the Internet Computer since 2021 and uses Caffeine as a paying customer, which made his experience all the more frustrating. According to his account, he attempted to build five different applications using the platform. Each started with promise, showing solid early designs and working first versions, but problems emerged once more detailed logic and state fixes were needed.

He said small corrections often spiralled into repeated loops and growing state issues, requiring hours of explanation without resolution. The result, he wrote, was lost time, credits and effort, with none of the projects reaching completion. While careful to avoid placing blame on individual users or the team behind Caffeine, he argued that many non-developers are now trying to build real on-chain backends, and without clearer guidance, frustration is difficult to avoid.

The concern, he added, goes beyond a single user experience. He said he has seen similar sentiments among other builders and investors on social media who support ICP, are willing to pay for tools like Caffeine, but feel uncertain about the platform’s limits and whether those limits are fully acknowledged.

Rather than calling for perfection, he argued for clearer communication about where Caffeine currently stands. He suggested that openly stating the product is still maturing, alongside clearer onboarding and guidance, would help set expectations and build trust. Among his specific suggestions were a mandatory foundation stage before building, improved onboarding for non-developers, and stronger, more visible support for paying users.

He also raised a reputational concern, noting that while ICP provides the underlying infrastructure, many users experience it through interfaces like Caffeine. If those interfaces create confusion, he warned, it can feel as though the network itself is falling short, even if the core technology remains strong.

Dominic Williams, founder of the Internet Computer, responded by acknowledging that Caffeine is still early in its development and that communication around its direction needs improvement. He outlined plans to move to a Phase 2 AI engine in the new year, shifting towards a more custom, agent-based framework designed specifically for the Internet Computer rather than relying on off-the-shelf solutions used by many traditional projects.

Williams argued that while pre-built agent frameworks can deliver quick gains, they also cap long-term potential. He said the aim with Caffeine is to build a custom system tailored to features such as Motoko and orthogonal persistence, which he believes will allow faster development, more complex applications and better ways for users to correct issues.

He also pointed to related upgrades, including changes to the Motoko language that would allow canisters to be defined across multiple files, and improvements that would enable work to be broken into parallel streams. Over time, he said, this approach should reduce unexpected changes, such as text being altered during generation, and make it easier for users to give precise instructions.

Williams accepted that these changes are complex and not always easy to explain, and said future Caffeine releases would include clearer ways for interested users to follow development progress. He also confirmed plans for a dedicated Caffeine blog to improve transparency, with further updates expected in early 2026.

Taken together, the exchange highlights a familiar tension in emerging developer tools. Enthusiasm for ambitious technology sits alongside the practical challenges faced by early users, particularly those without deep technical backgrounds. How effectively Caffeine bridges that gap may play an important role in shaping perceptions of the Internet Computer as the platform continues to evolve.