Detailed Analysis
A solo founder and consultant with ADHD has shared an open-source system built on Claude that effectively functions as an externalized executive brain, handling task prioritization, communications, legal work, and intelligence operations across six distinct role-based configurations. The user, who left corporate employment to run their own operation, describes the first month of solo work as nearly unmanageable due to ADHD-related memory and attention fragmentation — lost context between calls, forgotten follow-up rationale, and incomplete task threads. Rather than seeking behavioral fixes, they engineered a structured Claude-based workflow that delivers a daily action plan with energy tagging and time estimates by 8am, processes meeting transcripts into CRM-ready outputs, and generates outgoing communications written in the user's own voice — trained on four years of personal posts and direct messages.
The architectural detail that distinguishes this system from simpler AI assistant configurations is its shared memory layer across all six roles. A sales call debrief, a legal contract review, an OSINT investigation log, and a compliance memo all draw from and write to the same underlying context store, meaning information discovered in one domain can surface relevantly in another without manual cross-referencing by the user. This design directly addresses one of the core cognitive burdens ADHD imposes on knowledge workers: the overhead of context-switching and the lossy transfer of information between mental states. By offloading continuity to the system rather than relying on the user's working memory, the architecture treats the ADHD brain as a feature constraint to design around rather than a deficit to overcome.
The deliberate UX philosophy embedded in the system is notable for what it reveals about the emerging discipline of neurodivergent-centered AI tooling. The author explicitly codes against shame language, tracks effort metrics rather than outcome metrics (messages sent rather than replies received), and enforces a strict friction rule — if an action cannot be completed via copy-paste, click, or checkbox, it is excluded from the task surface entirely. These are not incidental preferences but represent a coherent design philosophy in which the interface's psychological safety is treated as a functional requirement, not an aesthetic one. The system fails if it generates guilt or demands cognitive overhead it was built to eliminate.
This account sits within a rapidly expanding category of user-built agentic deployments that repurpose large language models as persistent operational infrastructure rather than on-demand query tools. The six-role structure mirrors enterprise role-based access and delegation patterns, applied here at the individual level — a single person architecting what would traditionally require a paralegal, a chief of staff, an executive assistant, and an analyst. The open-source release of the configuration signals growing community interest in replicable, forkable AI operating systems for solo operators, a trend accelerating as the cost of building such systems drops to near zero for technically literate founders. Claude's capacity to hold complex role context, maintain voice consistency across writing tasks, and process structured documents like contracts and transcripts makes it particularly suited to this multi-domain, single-user deployment pattern.
The broader significance is that cases like this one demonstrate AI moving from productivity accelerant to cognitive prosthetic in the most literal sense — not making neurotypical workflows faster, but making previously unmanageable workflows viable. For the estimated 4-5% of adults with ADHD, many of whom are drawn to entrepreneurship, the gap between capability and executive function has historically been the difference between success and failure in solo operation. Systems that close that gap without demanding behavioral conformity represent a meaningful accessibility development in professional technology, one that commercial productivity software has largely failed to address.
Read original article →