DeskRoot | AI Workspaces

Your AI, set up for your work. Free. Guided setup for your business context, workflows and filing system. Plain files you own.

[DeskRoot | AI Workspaces @DeskRoot]

Your AI, set up for your work. You answer a guided interview once and end up with your business context, workflows and filing rules written down as plain files you own. Your own assistant reads them first. The DeskRoot foundation is free.

[DeskRoot | AI Workspaces @DeskRoot]

Most of what you re-explain to an AI assistant never changes: your customers, your pricing, which emails wait for your OK. Write it once in plain files you own, and every session starts there. DeskRoot's setup interview does the writing. Free.

https://t.co/vrwJeXVzvX

[DeskRoot | AI Workspaces @DeskRoot]

@traeracks Congrats on shipping. With payouts and contracts live, now's the time to write down the few rules Claude must never break, like never change payout logic without showing you the change first. One short file it reads before every session.

[DeskRoot | AI Workspaces @DeskRoot]

@Kainative The pipes-before-wall story is the best argument for this design. Worth one more step: keep those order rules as plain lines in your rules file, each with the job that taught it, so code can check every brief against them the way you already check citations.

[DeskRoot | AI Workspaces @DeskRoot]

@DemoGap Exactly, and none of it needs code. Write the check as plain steps: open this, click that, you should see this result. If those steps don't pass in front of you, it isn't done, whatever the assistant reports.

[DeskRoot | AI Workspaces @DeskRoot]

If an AI assistant tells you a task is done, ask what it checked. Better still, write down once what done means for the work you hand it: a report is done when the totals match the source, a customer email when you've OKed it. Then it checks before it says so.

[DeskRoot | AI Workspaces @DeskRoot]

@Mkandeger2 @dxiaolong Yes, a date and the reason on every line. Then a review every few months is quick: read the reasons, delete the ones that stopped being true, keep the rest. Without dates you end up re-reading the whole list just to find the old ones.

[DeskRoot | AI Workspaces @DeskRoot]

@CorneliusWrites Worth saving what it catches in a short checklist file it reads before each review. Then every new schematic gets checked against the things you've missed before, first, instead of the review starting cold each project.

[DeskRoot | AI Workspaces @DeskRoot]

@THE_REAL_CH3 @BluebriarArts Nice, that's the right behavior. Worth having it save which source you picked for each conflict in a short note it reads next time, so the same conflict doesn't come back to you every month.

[DeskRoot | AI Workspaces @DeskRoot]

@TazsBuilds Good one for the top. Worth one more line under it: when the reviewer flags something, it names the rule it's applying. Then you can tell a real rule break from a hunch, and a hunch that keeps coming up is your next rule.

[DeskRoot | AI Workspaces @DeskRoot]

@DemoGap That's the one to write first. Done means a check you can see pass yourself, not the assistant saying fixed. Put the check next to the rule, the exact steps you run, so every tool knows what done looks like before it starts and you're not taking its word.

[DeskRoot | AI Workspaces @DeskRoot]

A good test before you hand an AI assistant real work: could it tell you what your business does, which emails wait for your OK, and where finished work goes? If not, write those down once, in plain files your assistant reads first.

https://t.co/mBMBxsoh8S

[DeskRoot | AI Workspaces @DeskRoot]

@THE_REAL_CH3 @BluebriarArts That's a real win. Two things make it safer to keep: have it show the totals it pulled from each source file so you can tick them before you file, and have Claude write a short note on what each input is and what it checks. A format change then becomes a quick fix.

[DeskRoot | AI Workspaces @DeskRoot]

@dxiaolong When you pull them out, put a date on each one. Some no's are worth revisiting after a few months, and a dated list shows you which, so it stops the re-pitches without freezing every old decision for good.

[DeskRoot | AI Workspaces @DeskRoot]

@Mkandeger2 @dxiaolong It grows if you let it. Keep it to one line per no, each with the reason. When a reason stops being true, delete the line instead of arguing with it. Anything longer goes in its own note the list points to.

[DeskRoot | AI Workspaces @DeskRoot]

A good first file for any AI assistant at work: what it can finish on its own, and what comes back to you. Logins, money and anything a customer will see wait for your OK. Everything else, it finishes. Write that once and you stop re-deciding it every chat.

[DeskRoot | AI Workspaces @DeskRoot]

@dxiaolong @Mkandeger2 Saying no is the job. Worth keeping the no's in one file both tools read first, with a line on why. Otherwise when you swap between them, the one that wasn't there for the decision suggests the same feature again.

[DeskRoot | AI Workspaces @DeskRoot]

@thekenndubisi Worth writing that down so you only catch it once: a line in the instructions it reads first saying do every step you can yourself, and only hand me the ones that need my login, my money or my sign-off. Then the list it hands back is only what really needs you.

[DeskRoot | AI Workspaces @DeskRoot]

An assistant that writes well can still send the wrong thing if it doesn't know your rules: which emails wait for your OK, what you agreed last month, how you price. DeskRoot keeps those in plain markdown files your assistant checks first. Free.

https://t.co/R6nZzvIHOV