DoneThat vs No assistant

The short answer

Working without a desktop AI assistant means nothing interrupts you, nothing watches your screen, and nothing has an opinion about how your afternoon is going. For people who already notice their own drift and correct it, that is not a gap — it is a clean setup working as intended.

The case for an assistant is narrower than the category's marketing suggests. It is not that you cannot manage your own attention. It is that the afternoons which get away from you are, by definition, the ones you did not notice at the time.

The case for no assistant

No interruptions. If your best work happens in long uninterrupted blocks, anything that decides your attention is better spent elsewhere is a cost, and possibly a large one.

No capture. Nothing running, nothing processed, no vendor with any view of your work. This is the strongest argument on the page and it does not depend on anything else being true.

Nothing to tune. Every assistant needs configuring to the point where its interventions are useful rather than noise. That is real work, and until it is done the thing is a nuisance.

No dependency. No subscription, and nothing that can change under you or shut down.

Self-direction stays intact. There is a reasonable position that outsourcing attention management makes it weaker. We do not have evidence either way, but it is not a silly concern.

What self-monitoring actually costs

The catch with doing it yourself is structural rather than moral.

Noticing costs attention. Catching your own drift means running a background process that periodically asks "is this what I should be doing?" That process competes with the work for the same resource — and it is suspended precisely when you are most absorbed, which is when it would be most useful.

Hindsight is the usual outcome. Most people do notice the low-value afternoon. They notice at 6pm. The information arrives after it could have changed anything.

Patterns are invisible from inside. A single distracted afternoon is obvious. Three weeks where your stated priority got 15% of your attention is not, because no individual day looks wrong.

What Don actually does

Concretely, so it can be judged rather than believed:

• Nudges you to set goals, and can block time in your calendar for them.
• Pulls you back when you are on low-value work — while you can still fix it, not in tomorrow's summary.
• Asks how you are doing, so mood is tracked alongside the work rather than inferred from it.
• Encourages or judges, whichever character you pick.
• Reminds you of upcoming meetings.
• Suggests a break when you have been at it too long.

The reasonable objection is that this is a stream of interruptions dressed up as help. It is why how much Don does is something you set — you decide how it can help without it getting annoying. If you weight uninterrupted focus heavily, weight that setting heavily too.

On the capture requirement

The coaching only works because the system can see the work, so choosing an assistant means accepting capture. Precisely what that means: screenshots are never stored — processed in real time, deleted immediately, only the derived structured activity persists. Excluded apps are redacted on your machine before anything is sent. With your own Gemini or OpenAI-compatible model, raw activity never reaches our servers. The desktop app is open source.

If none of that is acceptable, no assistant is the coherent choice, and it is not a failure to understand the trade.

Where no assistant is the better choice

• You already catch your own drift in the moment. Some people do; if you are one, this page is not for you.
• Interruptions are expensive for you. Deep-focus work has a real re-entry cost.
• You cannot allow capture.
• Your work is singular. One task all day leaves nothing to drift from.
• You have tried assistants and turned them off. Usually a tuning problem, but a legitimate reason to stop.

Where DoneThat is the better choice

• You notice at 6pm, not at 2pm. Real-time beats retrospective for anything you want to change.
• You lose afternoons and cannot say to what.
• Your priorities drift over weeks in ways no single day reveals.
• You want mood tracked, so burnout risk shows up as a signal rather than a surprise.
• You want the record as well as the nudges — MCP and an API make your work history queryable as long-term memory.
• You want accountability without a manager. Follow others, share summaries, Slack and Discord, accountability groups — all opt-in.

How we compared these

There is nothing to cite: no assistant is the absence of one. Claims about DoneThat are first-hand — we build one of the two options, which is the bias to read this with.

We have not claimed that everyone needs a coach on their desktop. The narrow claim is that self-monitoring fails in a specific and predictable way — it stops working exactly when you are most absorbed — and that is the gap an assistant fills, if you want it filled.

Questions people ask about No assistant

Do I need an AI assistant on my desktop?

Not if you already notice drift while it is happening and correct it. Some people genuinely do. The case for one is narrower than the marketing suggests: it is for catching the afternoons that get away from you, which by definition are the ones you did not notice at the time.

Aren't AI nudges just another interruption?

They can be, and a badly tuned assistant is worse than none. It is why how much Don does is something you set — you decide how it can help without it getting annoying. If your working style depends on long uninterrupted blocks, weight that heavily.

What does a proactive coach do that self-discipline does not?

It notices from outside. Catching your own drift requires spending attention on watching yourself, which is attention not spent on the work — and it fails exactly when you are most absorbed. Don nudges you onto goals, blocks calendar time, flags low-value work while you can still fix it, checks your mood, reminds you of meetings and suggests breaks.

Can I use it without any capture?

Not meaningfully — the coaching is built on seeing the work. What you can control is what enters the record: screenshots are never stored, excluded apps are redacted on your machine before anything is sent, and your own LLM can process the raw activity so it never reaches our servers.