Claude chats rarely turn into finished projects because the plan lives inside the conversation, and the conversation ends. When you use Claude as a project manager, it needs a place outside the chat where actions, owners and status are written down and updated every session. Without that place, each new chat starts from zero and all the follow-through falls back on you.
You probably know the pattern. On Monday you spend forty minutes with Claude mapping out a product launch. You finish with a tidy twelve-step plan and feel properly organised. By Thursday you open a new chat, can't find the old thread, explain half the project again and get a slightly different plan. Two weeks later you have three plans in three chats, and the launch still hasn't moved.
I ran parts of my own business this way for months before I worked out what was going wrong. The thinking in those chats was good. What was missing was the dull, steady work a project manager does between conversations. This piece covers why that gap opens, the four causes behind it, and the simple file I now use to close it.
Why good plans disappear in a chat window
A chat window is built for conversation. You ask, Claude answers, and the thread scrolls on. That works well for thinking through a problem, and it works badly for holding a plan you need to come back to next week.
The longer a project runs, the more this matters. Older threads drop out of view, search across them is clumsy, and there is never one current version of the plan. Each new chat produces another plan, and each plan becomes another open loop in your head. If you've ever felt busy and stuck at the same time, those loose plans are part of the reason.
So the question shifts from "Is Claude good at planning?" to "Where does the plan live once the chat closes?" Most people never answer that second question, and that is where projects stall.
Four reasons your Claude projects stall
When I looked back at my own stalled projects, the same four causes showed up every time.
The plan has no permanent home
The plan exists only as a reply in a thread. You can't edit it, tick things off, or hand it to Claude next week without copying and pasting. A plan you can't update is a snapshot, and snapshots go out of date within days.
The steps are too vague to act on
Claude will happily suggest steps like "research competitors" or "finalise messaging". Those read well and give you nothing to do at 9am. Each item needs to be a physical, visible action, the kind described in GTD next actions, with a clear owner and a rough date.
Nobody tracks what is done or waiting
Real projects have items that are finished, items that are blocked, and items that sit with someone else. A chat has no way to show any of that. Delegated work quietly vanishes unless it goes onto something like a waiting for list that gets checked.
Every session starts without context
If Claude doesn't know your business, your priorities or what you agreed last time, you spend the first ten minutes of every chat rebuilding that picture. A Claude context file solves the background part. The live state of the project still needs its own home, and that is the job of actions.md.
What a project manager does that a chat doesn't
A good project manager keeps one list, updates it after every meeting, chases what is overdue and reviews the whole thing each week. None of that is clever work, and all of it is what keeps a project moving.
Claude can do every one of those jobs if you give it something to manage. That shift, from asking Claude questions to giving Claude a job, is the main idea behind using Claude as a project manager. It also explains why AI works best when it sits on top of a system you already trust. The system is the list, and Claude is the one who keeps it current.
The simplest version of that system I've found is one plain text file.
How to set up an actions.md file
You can set this up in under half an hour, and you will tune it as you go.
- Create a plain text file called actions.md in the project folder Claude can read.
- Split it into three priority headings: P0 for this week, P1 for this month and P2 for later.
- Write each item as a physical next action with an owner and a date.
- Add a Waiting section for anything that sits with someone else, with the date you asked.
- Start every Claude session by asking it to read actions.md and summarise what is open.
- End every session by asking Claude to update the file: tick off finished items, add new actions and move anything that is now waiting.
- Review the whole file once a week and cut anything that no longer matters.
Here is what a few lines look like in practice:
- P0: Email Sarah the revised launch date (Nishant, Tuesday)
- P0: Draft the welcome email copy (Claude drafts, Nishant approves, Wednesday)
- P1: Book two customer interviews for pricing feedback (Nishant, by the 20th)
- Waiting: Logo files from the designer (asked 3 March)
The end-of-session update is the step most people skip, and it is the one that matters most. If a meeting produced decisions, paste the notes in and ask Claude to pull out the actions, as shown in turning meeting notes into action items with Claude. Then ask it to write those actions straight into the file.
What changes when the plan lives in a file
Once the plan has a permanent home, a new chat stops being a fresh start. Claude reads the file, tells you what is overdue, and picks up exactly where you left off. You stop carrying the project in your head, which frees up attention for the work itself.
You also get an honest record. When something has sat in P0 for three weeks, you can see it, and you can decide to do it, delegate it or drop it. That weekly look fits neatly into a GTD weekly review if you already run one.
Frequently asked questions
Why does Claude give me a different plan every time I ask?
Each new chat starts without the previous plan in front of it, so Claude reasons from scratch using whatever you type that day. Small differences in your wording lead to different plans. Keeping the agreed plan in an actions.md file and asking Claude to read it first gives every session the same starting point.
What is an actions.md file?
An actions.md file is a plain text list of a project's open actions, grouped by priority, with an owner and date on each item. Claude reads it at the start of a session and updates it at the end. It acts as the single current version of the plan, outside any one chat.
Do I need Claude Projects or special software for this?
You need a place where Claude can read and update a file, which could be a project folder, a connected tool or simply pasting the file in. The format matters more than the software. A plain text list with clear priorities, owners and dates works in almost any setup you already use.
How is actions.md different from a Claude context file?
A context file describes your business: who you serve, how you work and what you care about, and it changes slowly. The actions.md file tracks the live state of your work: what is open, done or waiting this week. Claude needs both, one for background and one for current status.
Where to go from here
Your Claude chats stall because the plan has nowhere to live once the conversation ends. The thinking you do with Claude is usually sound. What it needs is a single file that holds the current plan, clear next actions with owners, and a habit of reading and updating that file every session. With that in place, Claude can act as a project manager that remembers, chases and keeps your projects moving between conversations.
The actions.md file works best when it sits on top of a wider way of capturing, prioritising and reviewing your work. If you want that whole foundation, the NP System free course walks you through the full system that this file plugs into, so Claude has something solid to manage.
