Nexthen
From Scattered Intent to the Next Right Action
Product DesignerA to-do list with an AI assistantIn daily use
AIProduct DesignUXUIMCP
01. What problem it solves
Nexthen exists because a to-do list fails at the moment it matters most. Writing a task down is easy, but the list does nothing when a person cannot get themselves to start, cannot decide what to do first, or cannot bring themselves to add the thing at all. Sorting the list later becomes one more task and one more place to get stuck.
Three tasks, no first one
A thing too heavy to write down
Nexthen is a plain board of categories, tasks and subtasks, and the person talks to it through the AI assistant they already use, such as Claude or ChatGPT. When they mention something to do, the assistant files it straight into the right category. When they want help, they ask in the same chat, and the assistant answers with a method that fits how they work. The methods come from a library of books on productivity, philosophy and psychology, for example breaking a task down to the next physical action, or starting on one small piece instead of waiting until it feels easier.
02. The work
Nexthen is three parts that work together. The board at nexthen.now holds the tasks, and the nexthen skill tells the assistant in the person's chat with Claude or ChatGPT how to work with the board, with the working method, the book knowledge and the person's profile behind it. The assistant reaches the board through a connector. On a desktop, the board can be installed as an app.
For example, someone writes in the chat that they keep putting off their subscriptions and asks to be walked through it. The task Review monthly subscriptions already sits under Personal Finances with three steps, from finding the payments that repeat every month to cancelling the ones no longer used. The assistant names the first step, to open the banking app and look through last month's card payments for anything that repeats. When the person answers that they found seven, it ticks that step on the board and names the next one, to write the seven down with what each costs a month. The task stays open until the unused subscriptions are cancelled.
03. My role
Nexthen is a to-do list with an AI assistant that I designed and built, and it is the system I run my own tasks on. I designed the product, its interface and its design system with dark and light themes, and how the assistant behaves when someone asks it for help.
I decided that Nexthen stays a board for one person, without team features or extra statuses, so that it never turns into a project-management tool for teams. It also has no check-ins, streaks or counters, because generic productivity advice and added helpers are what make people stop using a to-do app. The person keeps control over how much help they get, from Light to Guided, and over what the assistant knows about them.
04. Key decisions
01
Nexthen runs on the AI subscription the person already has
I first planned Nexthen as a paid subscription with its own AI model and a named helper for the person to talk to, and I changed it so that Nexthen is free to use with the person's own Claude or ChatGPT, because that assistant already knows the person from everything they have told it. The product holds the board and the profile, and every bit of AI work runs in the person's own subscription. On a phone, Claude or ChatGPT reads the full nexthen skill and the book library from Nexthen itself, so the person gets the same help there as on the computer.
02
Brain Dump stays the person's own lane
Brain Dump is the lane where the person writes things down for themselves. Another early plan routed everything the person mentions to the assistant into Brain Dump to be sorted later, and I rejected it, because the person would then find their own notes mixed with tasks they never wrote and still have to move each one. So the assistant decides where every task goes by reading the category names and the tasks already in them, and when two categories fit, it asks one short question with both names and remembers the answer.
03
The book methods became part of how the assistant works with the person every day
Nexthen first had a Guidance Library on the board, with written summaries of the methods and a Guidance Engine that applied them, and I removed both, because the knowledge belongs where the conversation happens, on the assistant's side.
The library behind the assistant holds many books because people work in different ways. One person writes everything down and works through it as Getting Things Done teaches, while another keeps only a few things on the list as Essentialism argues. The assistant knows from the person's profile how they work and draws on the books that match it, and every book is prepared so that the assistant finds the right passage quickly.
04
The assistant takes on a task the way a real assistant would
The person moves a task that needs research or a written answer into the AI Assistant lane, and the assistant does the work and saves its answer there, with links to its sources when it used outside information. Saving the answer completes the task and keeps the answer in Saved Responses, a library the person can search later. The assistant can use every action the board has, so it also creates, sorts, schedules and restores tasks, and when a new task fits none of the categories, it names the category it would create and creates it unless the person says no.
05
Nexthen is built to work for people with ADHD and autism
When someone cannot bring themselves to add a task, a single word is enough, and the assistant writes the full task, files it and confirms in one line without asking when, why or how. When the person asks what to do now, the assistant asks at most two short questions, one at a time, about what it cannot tell from the profile, such as the time and energy they have, then names one task and says in plain words why that one.
05. How it works
From a sentence in the chat to the board
Every request in Nexthen runs the same way, from a sentence in the chat to a change on the board. Before the assistant acts, it loads three things, the nexthen skill with the working method, the person's profile and the current board. Someone writes in one message that the launch is on Thursday and they should email the client about the footer links today, that the car needs its inspection before the end of the month, that their mom's birthday is on the 12th and they need a gift, and that the dry cleaning is waiting again. The assistant sorts all four against the board. The footer links email is marked for Today's Wins, the up to three tasks the person keeps as the day's focus, because the person wants to send it today, before the launch on Thursday. The car inspection goes into Errands & Appointments, and the birthday goes into Home with the gift as its first step, while the dry cleaning is already on the board and is not added twice. It writes each change through the connector with the version of the board it read, the server checks the change and saves it, and every device that has Nexthen open shows the new tasks a moment later.
The difference from other to-do apps
In an ordinary to-do app, the person does all of this alone, from deciding where a task goes to choosing what comes first and breaking a large task into steps. In Nexthen the assistant does that work with the person, because it has the whole board and the profile in front of it and writes the result straight back to the board.
The profile
The profile is what makes each answer fit the person. It starts with the questionnaire the person answers the first time, with questions such as when their head feels clearest and what they say to themselves when they put something off, and with up to three goals and the one that comes first. The answers set which hours the assistant keeps for hard work, how much it decides for the person, which way of starting it offers first and the tone it uses. After that the assistant keeps watching what the person says, adds and finishes, and writes the patterns it notices into About you, a part of the profile only the assistant writes and the person can always read. The person's own word beats any observation, and whatever the assistant learns changes its next decision.
The method and the books
The books behind the method are split along their own chapters and sections into short passages, and when the method calls for a book, the assistant searches the library with the person's actual question, reads the exact passages that come back and answers from that text. Summaries of the books and a map of where they disagree tell it where to look first and which books to read side by side. A new book reaches every assistant through a library release on the server, and the app itself stays the same.
One board on the server
The board on the server is the one record every change is checked against. Each change carries the version of the board it was based on, and the server saves it only when nothing has changed in between and the change keeps the board's structure intact, so two assistants or two devices never overwrite each other. A change that fails writes nothing, the assistant reads the board again before it tries once more, and before it tells the person that something was added, it reads the board back to confirm it. When something is deleted, the server hands back a copy that restores it exactly.
06. Evidence
1,400+confirmed changes on the board
931automated tests pass
11 of 11backups, every checksum matching
4,825passages in 35 works
Daily use
The board behind my own tasks has recorded more than 1,400 confirmed changes, made by me, by the assistants working on it and by the server's own scheduling rules, and each of them passed the server's checks before it was saved.
Every release is checked against the live board
Every release is checked against the live board as soon as it is switched on, and a release that fails a check rolls itself back. On the latest release the board kept the same version and the same fingerprint, a checksum of the whole board, before and after the switch, a backup was taken first, and after the server restarted, a final backup was taken and restored. One candidate rolled back because Chrome on the computer showed a gap of 36.65 pixels where the design specifies 20, caused by a scrollbar, and the corrected release stayed live once the gap measured 20. The latest release passed all 931 automated tests.
Backups that restore
The board is backed up every day in encrypted form and copied to a storage drive at home, and the first copy of 11 backups arrived with every checksum matching. Restores have been tested from the server, from the copy at home and with the key from a password-protected recovery package, and each one returned a complete, working board, while a monthly test restores a backup in a sandbox.
The book library
The book library holds 35 works in 4,825 passages. Its verification reads sampled passages through the same connection the assistants use, turns away anonymous and invalid requests, and confirms that the board did not change during the check, both before and after the service restarts.
Sign-in and accessibility
Every release confirms that the board cannot be read without signing in and that an assistant without valid access is turned away, and the sign-in tests cover revoking an assistant's access. An accessibility review of the seven main screens left no serious or critical issue open.
07. What I learned
A rule meant to protect the person must never stand above their own words
To stop the assistant from labeling anyone, the server rejected a list of words in the text the assistant saved to the board. In an urgent moment that list rejected a task the assistant was filing for me, about myself and in my own words, and I decided the same day to remove the whole list. Nexthen now saves whatever the person writes and keeps a record of who wrote it, because a tool that corrects how someone describes their own life has stopped helping them.
Once a product does what I need, the next work is making change safe
When the board had everything I wanted for my own tasks, the biggest risk was that each new change could break something that already worked, from the layout to the saved data. So before the next round of features I built a lab that checks the interface against fixed reference images at three screen widths, together with the release process that guards the live board. The features kept coming after that, from the questionnaire and About you to the book library that every assistant reaches from the server, and the board's data has come through every release since then unchanged. Time spent making change safe is what lets a product keep growing once it already works.