Felix Rieseberg, the engineering lead for Claude Cowork and Claude Code Desktop at Anthropic, described three personal projects on Lenny’s Newsletter’s “How I AI” podcast that together illustrate how the ceiling on self-built software has dropped. The examples are concrete enough to be instructive on their own terms.
The first project started with a house move. Rieseberg wanted to build a 3D floor planner showing his actual furniture. Rather than measuring and entering dimensions manually, he pointed Claude at years of email receipts and order confirmations. Claude parsed purchase records, extracted dimensions and item names, and assembled a personalized inventory that fed directly into the floor planner. The insight here is not that AI is clever; it is that most people already have a structured personal database sitting in their inbox and have not thought to use it that way.
The second project extended the same principle to live dashboards. Claude Cowork supports what Rieseberg calls live artifacts: outputs that refresh in real time by pulling from connected services such as Gmail, Google Calendar, Spotify, or Notion. He built a personal dashboard that updates throughout the day without manual input. The practical implication is that personal reporting, daily briefings, and status summaries become self-maintaining once the connections are authorized.
The third pattern is really a meta-lesson across the other two. Rieseberg describes going one abstraction layer up repeatedly: he was entering furniture data himself, then told Claude to figure out the furniture list, then told Claude to find the furniture in his email. Each step removes a bottleneck that would otherwise require the user to do repetitive work. The pattern applies to any domain where someone is doing structured but tedious extraction.
Taken together, the three examples make a real case that the bar for building personal software has fallen. Workflows that previously required a developer, a database administrator, or a no-code platform with significant setup time can now be assembled in a single session with a capable AI tool and access to personal data sources.
The skepticism worth applying here is direct. Rieseberg is not a neutral user of Claude Cowork; he built it. His demonstrated workflows are also excellent product marketing for the features he shipped, including live artifacts and email connectivity. That does not make the examples false, but it does mean the implied audience for “anyone can do this” is narrower than the framing suggests. Rieseberg has a decade of engineering background that shapes which problems he recognizes as solvable and how quickly he structures prompts to solve them. The floor planner project required recognizing that email receipts were a data source, understanding what a floor planner needs as input, and knowing how to connect those two things. That recognition is not evenly distributed.
What the examples do establish is a set of concrete affordances: personal email as a queryable inventory, connected apps as live data sources, iterative abstraction as a workflow methodology. These are real capabilities, not aspirational ones.
For founders building consumer or prosumer products, the question Rieseberg’s projects surface is which “personal software” use cases your users would build if they had a Claude Cowork-class tool connected to their data. Those are the integrations worth examining first, because they represent the workflows your users currently cannot do without a developer, a spreadsheet, and patience.
Posted by Lenny Rachitsky in Lenny’s Newsletter on 2026-05-26.
