
Hey everyone! Haven't posted here for a while... Feels a bit awkward. Over the last ~6 months I've been working hard on a new version of Harmonica. It's been a looong sprint and I feel burnt out, but also satisfied with the end result. Considering that I had never written any code until December 2025, it feels a bit surreal to me. Building my own product with Opus 4.5+ has been the quickest education I've ever received in my life, a real speedrun. Of course if wouldn't be possible if I started from scratch, so I'm super grateful to our developers who laid out the important foundations in the prior versions. Since its inception at Newspeak House I wanted Harmonica to execute facilitation methods from well-known "traditions" such as Wardley Mapping, Team Topologies, Cynefin, Sociocracy 3.0, Agile, etc. I think of these protocols for "thinking together" as distilled sensemaking wisdom, but I've always struggled with deploying them in a community or with clients. You can make a Wardley map yourself but unless other stakeholders are interested in this kind of activity, it's going to be YOUR map, created in single-player mode. Without collective input it's going to be biased and lack the domain knowledge that's critical for making strategic decisions. But to collect that input you need to understand the methodology, explain it to others and find time for the workshops. The friction is too high, so the protocol gets deferred more often than not. After dozens of interviews with our ICPs -- who we broadly identified as change managers, org designers, governance leads and consultants -- we knew there's enormous value in "collecting deeper insights at scale" but their workflows are sophisticated and surveys, even if very adaptive and multi-lingual thanks to LLM, can have a limited impact. Harmonica 0.4 is finally solving this problem with what I provisionally call "a runtime for facilitation" or "a methodology execution engine". I'm still searching for a better term, but the most important thing is that it's not "a mere AI interviewer" anymore. It's probably a worse interviewer than Listen, Juno or Genway. But now it's capable of something they're not designed to do, and whiteboards like Miro are not designed to do either: execute facilitation protocols, following the best practice in context engineering and AI evaluation. Read the full story in our blog: blog.harmonica.chat/runtime-…













