I code with ADHD by moving what my brain drops into the tools around it. Claude Code, my AI coding tool, reads a notes file called CLAUDE.md at each start, and small scripts called hooks show where I left off. My brain stays the same, but my setup does the remembering and holds the finish line, so I ship. A plain note in your project folder works too, and my hooks just make sure I write it and read it when I forget.
On 06/22/2026 I wrote a hook that greets me with three lines every time I open Claude Code. I built it because my picture of the work did not survive the night. This is the card, with the real text swapped for placeholders.
↩ were: <what I was doing>
→ next: <the open loop>
✓ last session shipped: <receipts>The first line says where I stopped. The second line is my next step, taken from the open loops in my last handoff note. An open loop is a job I started and have not closed. The third line lists what the last session actually shipped, from my commits and the goals I closed.
My ADHD brain while coding, and what Claude Code carries for each part
What Claude carriesCLAUDE.md, a notes file that Claude Code reads at the start of every session
I lose the second half of a plan while typing the first half. More on working memory →
What Claude carriesWhen a session ends, a hook (a small script that runs at a set moment) writes a handoff note. It says where I was, what I meant to do and what is still open. When the next session starts, another hook shows that note as a 3-line resume card.
Every Claude Code session starts with a blank short-term memory, and so does my head the next morning. More on losing my place between sessions →
What Claude carriesThe 'next' line of the resume card, plus one line on my screen that shows my current goal
A blank session gives me nothing to grab. More on getting started →
What Claude carriesA saved note before every switch. A hook also warns me when my prompt shares under 12% of its words with my current goal, a sign that I may have drifted.
A switch without a written note means the next session starts cold. More on context switching →
What Claude carriesA written finish line for each goal, and a dated note to re-check every fix
I start well and close badly. More on finishing (open loops) →
What Claude carriesWhen a session ends, a hook adds that session's commits and closed goals to a log. It writes nothing if the session was empty.
I forget what I shipped yesterday. More on forgetting my own progress →
What Claude carriesA bar at the bottom of my screen with the Pacific time and the time since my last commit (my last saved change)
I cannot feel time pass while I code, so I need to see it. More on time blindness →
Selected
Losing my place between sessions
Every Claude Code session starts with a blank short-term memory, and so does my head the next morning.
Context switching
A switch without a written note means the next session starts cold.
Each thing my brain drops while I code has one tool in my setup that holds it instead.
Source: my own setup, from my Claude Code hooks and notes (built 05/22/2026 to 10/02/2026) · 10/10/2026
This is my lived experience as a developer with ADHD. It is not medical advice, and nothing here replaces a doctor or therapist.
Can you be a good programmer with ADHD?
Yes, I think so, and I am trying to be one in public. I am a self-employed engineer who builds with Claude Code every day.
The research agrees that it cuts both ways. A 2024 case study at ICSE (the International Conference on Software Engineering) looked at engineers with ADHD [5]. It found real struggles with organising tasks and estimating how long they take.
It also found strengths, like creative skill and doing well on puzzles. That matches my week pretty closely.
So I stopped asking whether I can code with ADHD. I ask which parts my brain drops, and what can hold them instead.
Why is coding so hard with an ADHD brain?
Coding leans hard on executive function, and executive function is the part of my brain that ADHD makes unreliable.
NIMH (the US National Institute of Mental Health) calls ADHD a neurodevelopmental disorder, which means it has to do with how the brain develops. NIMH also notes that symptoms often continue into adulthood [1]. CHADD (a US nonprofit for people with ADHD) says executive function problems hurt the ability “to begin, work on and complete tasks” [2].
Four things break.
Working memory is the first leak. It is the brain’s scratch pad, where I hold a plan while I work on it. I lose the second half of a plan while I type the first half.
A 2005 meta-analysis found working memory problems in children with ADHD [3]. The effects were larger on spatial tasks, which are about where things sit, than on verbal tasks. That study was on kids, so I treat it as a direction for my own case. I do not treat it as a number about me.
Context switching is the second. When I jump to another task, I almost never stop to save my place, because in the moment I feel sure I will remember. I come back to my own code as a stranger.
Starting is the third. It is sneaky. An empty prompt box gives me nothing to grab. So I drift into tuning my tools instead of starting the thing I sat down to build.
Losing my place between sessions is the worst. Claude Code has the same problem, and that surprised me. Its own docs say each session begins with a fresh context window, which is everything the AI holds in mind while it works [6]. So I had two forgetful workers on the job, and neither one kept notes.
My own logs showed me what this costs. My setup keeps a log of every skill it runs. A skill is a saved set of instructions that Claude Code loads when a task needs it. On 06/04/2026 I read the last 14 days of that log.
The one skill that does my paid work ran 23 times in those two weeks. The top eight skills that tune my own setup ran 352 times between them. I also had eight or more big jobs open, and none of them had closed.
I was busy. I was not finishing.
If you lose the plan every time you close the laptop, start with my free ADHD CLAUDE.md config. Copy it into your project tonight, and Claude Code will read it at the start of every session.
What does Claude Code actually carry for me?
Claude Code carries my place in the work and my finish lines, so my brain does not have to hold them alone.
The idea is old. Russell Barkley, the ADHD researcher, has a rule about this in his book on adult ADHD [4]. The rule says to write key information down outside your head, where you can see it.
My reading of it is simple. If it lives only in my head, it does not exist. So I built my setup around one test. When two designs compete, I pick the one that asks me to remember less.
CLAUDE.md is the base layer. Claude Code reads it at the start of every session [6]. The docs suggest keeping it to facts every session needs, like the commands that build the project and its coding rules [6].
Hooks changed my days. Claude Code runs each hook at a fixed moment, like session start or session end [6]. It runs every time, whether I remember it or not. That matters a lot.
I learned this the hard way. In May I added a written rule to the notes that Claude Code reads. It told Claude Code to stop asking me to do steps it could do itself. Within eight days I caught the same failure twice.
The rule was there, but nobody was paying attention to it. Claude Code’s docs say this directly. CLAUDE.md is only guidance that Claude Code reads. To block an action, you need a hook [6].
So now, when a failure keeps coming back, I move it out of written rules and into a script.
My own notes show that written rules alone did not hold. They do not prove that hooks always work, because I only measured short stretches of time. I bet on hooks, and I keep watching them.
Then skills. I use one skill that reads my git history, so I stop pretending I did nothing this week.
How do I pick up where I left off after a break?
I let the last session tell me. I do not try to remember.
When a session ends, a script saves a short handoff note. It says where to resume and what I meant to do, and it lists the open loops.
A second script counts what I actually shipped in that session, meaning my commits and the goals I closed. If the session was empty, it writes nothing. No fake wins.
When the next session starts, a third script reads both notes. It prints the three-line card from the top of this post before I type a single word. On a clean first session it stays quiet.
One Claude Code session, start to finish, and what carries each step
Step 1 of 7
Open Claude Code
MeI open a terminal and start a session.
What carries itClaude Code reads my CLAUDE.md notes file at the start (Claude Code docs, checked 10/10/2026).
Step 2 of 7
Read the resume card
MeI read three lines before I type anything.
What carries itA hook that runs when the session starts prints a 3-line card. It says where I was, what is next and what the last session shipped. It prints nothing on a clean first session.
Step 3 of 7
Set one finish line
MeI write what done means for this session.
What carries ita goal with a written 'done when' line, shown in one line on my screen
Step 4 of 7
Work
MeI build, test and wander.
What carries ita hook checks each prompt I send. It warns me when the prompt shares under 12% of its words with my current goal.
Step 5 of 7
Track tasks
MeI check off steps.
What carries ita live task list, also copied to a file that only grows. On 07/06/2026 a list of 15 tasks vanished mid-session, and the file kept the record.
Step 6 of 7
Close
MeI stop, sometimes abruptly.
What carries itA hook that runs when the session ends writes the handoff note and logs what I shipped. It writes nothing for an empty session.
Step 7 of 7
Next session
MeI come back, often the next morning.
What carries itStep 2 reads what step 6 wrote, so the loop closes.
A session in my setup starts by reading what the last one wrote and ends by writing what the next one will read.
Source: my own hooks (resume card and session log built 06/22/2026; a lock so only one session can take each handoff added 06/30/2026) · 10/10/2026
The “next” line is the part I did not expect to love. It solves starting, because it hands me a first step that already exists. A version of me wrote it the night before, while he still had the whole picture in his head. I do not have to invent the first move cold.
I also copy my live task list to a plain file that only grows. I started that after a task list vanished in the middle of a session. The work survived in the file. Now the file is the source of truth, and the list on my screen is only a display.
No hooks yet? My ADHD prompts for Claude post has a checkpoint prompt and a re-entry prompt. For me, they are the only two that survive interruptions.
How do I finish things instead of starting new ones?
I write the finish line down before I start, and I put a date on the check. Then a file holds the loop open so my head does not have to.
Every goal in my setup gets a “done when” line. It is one sentence that says what finished looks like, and it stays in one line on my screen while I work. A hook also warns me when a prompt shares under 12% of its words with that goal, a sign that I may have drifted.
Claude Code follows the same rule in my setup. Every promise it makes needs a due date and a check that anyone can run. It cannot call anything done without proof.
When I ship a fix, I save a dated note to check it again later. That note keeps the loop open, so I do not have to remember to come back.
None of this cures the scatter. I still scatter, and I think that is fine. The difference is that the scatter is safe now, because the loops live in files.
Where should an ADHD developer start?
Start with the one failure that costs you the most, and fix only that. My tool stack post says the same thing. Trying everything at once tends to backfire. I think that is because every new tool is one more thing my brain has to remember to open.
If losing your place is the worst part, start with a CLAUDE.md file. My free ADHD config file at /guide/adhd is the one I would copy first. If starting is the worst part, try the ADHD prompts for Claude that I linked above. If you keep finishing nothing, look at my energy-based planning system, which fits on one screen.
Hooks come later, once you know which failure keeps coming back. My written-rule story above is the reason I think that order works.
If your plans fall apart the day your energy drops, you can start from the same one-screen Notion templates I built for those days. Get the ADHD Engineer’s Productivity System for $19 and set up the Daily Driver Dashboard tonight.
Sources
- National Institute of Mental Health, “Attention-Deficit/Hyperactivity Disorder.” https://www.nimh.nih.gov/health/topics/attention-deficit-hyperactivity-disorder-adhd (checked 10/10/2026)
- CHADD, “Executive Function Skills.” https://chadd.org/about-adhd/executive-function-skills/ (checked 10/10/2026)
- R. Martinussen et al. (2005), “A meta-analysis of working memory impairments in children with attention-deficit/hyperactivity disorder.” Journal of the American Academy of Child and Adolescent Psychiatry 44(4), 377 to 384. https://pubmed.ncbi.nlm.nih.gov/15782085/ (checked 10/10/2026)
- Russell A. Barkley with Christine M. Benton, Taking Charge of Adult ADHD, second edition (Guilford, 2021), chapter 19, “Rule 4. Externalize Key Information.” https://www.guilford.com/books/Taking-Charge-of-Adult-ADHD/Russell-Barkley/9781462546855/contents (checked 10/10/2026)
- G. Liebel et al. (2024), “Challenges, Strengths, and Strategies of Software Engineers with ADHD. A Case Study.” ICSE 2024, Software Engineering in Society track. https://conf.researchr.org/details/icse-2024/icse-2024-software-engineering-in-society/11/Challenges-Strengths-and-Strategies-of-Software-Engineers-with-ADHD-A-Case-Study (checked 10/10/2026)
- Anthropic, Claude Code docs, “How Claude remembers your project” and “Hooks reference.” https://code.claude.com/docs/en/memory and https://code.claude.com/docs/en/hooks (checked 10/10/2026)
Grab my free ADHD CLAUDE.md config at /guide/adhd, and I will send you my neurodivergent systems notes as I write them.
Keep reading