An AI support system for neurodivergent freelance work
How I separated strategy, weekly execution and cross-project continuity so I do not have to rebuild the whole map every time I return to work.
I work independently across consulting, research, writing and digital projects. I designed a small set of AI collaborators and a shared Systems Office around the places where that work repeatedly became difficult: context disappeared, ideas multiplied faster than I could act on them, and ordinary planning systems assumed a steadier brain and a more orderly week than the ones available.
The problem I was trying to solve
Executive-function difficulty does not always look like forgetting what to do. For me, it often looks like having to reconstruct the whole landscape before I can decide what matters now.
I have worked independently for years. That gives me freedom, but there is no team quietly holding the larger map while I move among client work, research, writing, websites and whatever complicated new thing has captured my attention.
After an interruption, I may remember the project but lose its working state: why a decision was made, which version is current, what remains unresolved and where I had intended to return. A task list preserves the noun. It rarely preserves enough of the thinking to let me re-enter the work.
My capacity also varies. On one day I can hold a large conceptual problem in my head for hours. On another, reading the document required to begin may consume the available energy. The most important task is not automatically the task that can move today.
Then there are the ideas. I generate them faster than I can use them. Capturing an idea is useful; allowing every captured idea to present itself as a new commitment is how Tuesday ends up governed by a coup.
This is one neurodivergent person’s system. Autism and ADHD are not a shared workflow specification. The transferable method is to notice where work repeatedly fails in real life and design support for those moments.
How the system works
The system separates forms of support that are easily blurred when one general AI conversation tries to do everything.
| Part of the system | What it does |
|---|---|
| Coach Tessa | Holds career and business strategy, the professional portfolio and the larger direction across interruptions. She helps me compare opportunities and decide what deserves priority. |
| Sprint Tessa | Turns accepted priorities into a small, realistic week. She helps choose work that fits today’s capacity, cognitive mode and practical conditions, and records useful stopping points. |
| Chief of Staff Sol | Maintains cross-project continuity: why an idea exists, where it came from, what it changes and which other work it may affect. Sol also helps decide where new material belongs. |
| The Co-Intelligence Systems Office | Provides shared memory for the collaborators: an organisational map, specialist returns, decisions, workstream state, professional terrain and re-entry information. |
The names make the roles easy for me to recognise. They are optional. The important part is the division of labour.
A conversation that helps with long-term strategy should not silently decide today’s workload. A research return should not become a commitment because it is interesting. The system keeps capture, review, decision and assignment separate.
From recurring barrier to design response
| What kept happening | What was missing | Design response |
|---|---|---|
| Work spread across separate projects and conversations | Someone or something holding the relationships among them | Bounded specialist roles plus one cross-project continuity partner |
| Returning after interruption required extensive reconstruction | The last meaningful state, reasoning and route back | Current-state records, short returns and re-entry briefs |
| Every promising idea competed with existing work | A boundary between intellectual value and a claim on my time | Store candidate work without activating it |
| The prestigious priority did not fit the brain available that day | A choice among approved work based on cognitive mode | Plan by capacity and type of attention as well as project |
| Organisation systems became another job to maintain | A reason for each piece of administration to exist | Save only changes another role needs or that would be costly to reconstruct |
What happens when a new idea appears
Suppose a scheduled research search finds a paper that sparks a promising connection.
- The relevant specialist conversation examines the paper and develops the connection. It can submit a short return to the Systems Office if the finding has consequences outside that workstream.
- Chief of Staff Sol checks whether the connection changes another project’s assumptions, decisions or external outputs. Most specialist detail remains local.
- Coach Tessa sees it only when it affects professional direction, an opportunity, a commitment or the wider portfolio.
- Sprint Tessa receives it only after I have accepted it as work for the week.
Interesting has no automatic right of entry to Tuesday.
This boundary matters because AI can generate and elaborate possible work far faster than one human can complete it. A system that treats every useful output as an action has not solved overload. It has industrialised it.
Returning after interruption
Interruption is part of the design, not evidence that the system failed.
When I return, the first task may be to recover the current state rather than finish anything. The system can retrieve the last meaningful stopping point, decisions already made, unresolved questions, dependencies and the next available move. If priorities have changed, Coach Tessa recovers the larger map before Sprint Tessa plans the week.
Restoring access to a piece of work does not create an obligation to finish or activate it.
That distinction removes a surprisingly large amount of background guilt. Something can remain valuable, retrievable and inactive.
Why this helps me
Context can live outside my working memory. I do not need to preserve every relationship among projects while concentrating on one of them.
Different levels of work remain different. A ten-minute administrative task does not sit beside a professional transition pretending to be the same kind of priority.
Capacity affects selection without rewriting strategy. A low-capacity day can still be useful without declaring the larger work abandoned.
Ideas can be preserved without becoming assignments. Captured, developed, accepted and active are different states.
The system also adapts its manner. Strategic thinking can be exploratory. Daily support is brief and concrete. Low-capacity interaction reduces idea density and avoids turning an offload into another planning session.
Start with a much smaller version
My current arrangement grew around several years of accumulated work. It is evidence that the method can expand, not a sensible starter pack.
A useful first experiment may contain:
- one central conversation that remembers important decisions and helps review what matters;
- one lighter working conversation, only if daily action genuinely needs a different pace or type of context;
- a short handover containing what changed, what was decided, what remains open and the next meaningful move;
- one or two dependable reminders or records in the ordinary tools already used.
Begin with a real incident rather than an ideal routine. This prompt is enough to start:
Help me design a small AI-supported system around how I actually work. Start with one recent occasion when something important was forgotten, lost, delayed or repeatedly reconstructed. Do not design any AI roles yet. Help me identify what failed, at what moment, what information or action was missing and whether AI is a suitable part of the solution.
Test the smallest useful arrangement for two weeks. Notice what you use, what you avoid, what becomes easier and what creates fresh maintenance. Merge or remove anything that has not earned its upkeep.
Boundaries that matter
- Calendars, formal records and specialist systems remain authoritative for anything safety-critical, legal, financial, medical or otherwise consequential.
- An AI summary is not stronger evidence than the source material behind it.
- Material appearing in a shared system does not become a decision, commitment or weekly assignment by itself.
- Sensitive personal or client information needs deliberate privacy choices.
- Every additional role creates routing and maintenance costs. Complexity has to reduce more work than it creates.
The part worth copying
The useful idea is not my collection of named collaborators. It is the decision to design support around the work and the brain that already exist, including interruption, uneven capacity, intense curiosity and an entirely reasonable dislike of maintaining systems for their own sake.