TODO for AI

This is a great concept! 🚀 An AI agent that doesn’t just manage tasks but actually takes action is a big step forward. I’m especially interested in how it handles real-world workflows and learns from user feedback. Exciting to see AI moving from assistants to action-takers! 👏

Love that this works from a simple TODO list instead of needing a perfect prompt — writing 'reply to the 3 open support tickets' and having an agent pick it up in my own browser feels really practical for a small team. The way it learns from what you accept and reject is a nice touch too!

Hi Fazier! I'm Marcell, maker of TODO for AI. It's a TODO list where the items get done by AI agents: drop a task, an agent logs into your tools (browser, email, code, calendar, CRM) and does the work in your own accounts, then shows you the result. It learns your business first so it knows what's worth doing next. Free tier available — happy to answer questions here.

This is an amazing App. This is very Useful 🥉

This is an amazing App. This is very Useful 🥉

This is an amazing App and very Useful.

Turning a plain TODO list into something agents actually execute is a clear idea. Running the work in the user's own browser so it reuses existing accounts is a sensible design choice. I would be curious how it handles tasks that need review before sending, like emails or posts.

The part that stands out to me is that it runs in your own browser with your own accounts - most agent tools ask you to hand credentials to their cloud, which is a hard sell for a small team. One thing I would want to know before trusting it with my list: does it show its plan before acting on a TODO, or does it just go? A quick approve/reject step per task would make the learning-from-feedback loop feel a lot safer.

Lo más valioso de esta idea es que elimina la fricción de redactar el prompt perfecto: escribir «responde a los 3 tickets abiertos de soporte» ya es suficiente para que el agente actúe. Usar cuentas propias dentro del navegador también es una decisión de diseño muy práctica para el trabajo administrativo, porque evita el riesgo de compartir credenciales. Me gustaría saber cómo maneja la verificación antes de ejecutar acciones sensibles, como enviar correos o publicar cambios. Muy buen lanzamiento.

The idea of ​​transforming a regular to-do list into actual tasks for employees is excellent.

Working from a plain TODO list instead of a prompt is a neat idea, especially since it runs in my own browser with the accounts I already use. Curious to see it start proposing TODOs on its own.

Amazing work! 🚀 Excited to see what you’re building in the AI space. Keep pushing the boundaries of AI! 👏

Running the agent in my own browser on my own accounts is a smart trust model - no handing over OAuth tokens to a third-party server. The accept/reject learning loop is the part I'd watch: does it handle ambiguous TODOs like "post this week's update" well, or does it still need a fairly spec'd task to be reliable? Either way, for founders drowning in small ops work this framing (TODO list that closes itself) is compelling.

Running the agent in your own browser with your existing logged-in accounts is a smart way to avoid the usual integration setup. How do you let users review or undo actions before something like a LinkedIn post goes out?

Running the agent in my own browser so it can reuse existing logged-in accounts is a clear differentiator. How do you handle risky TODOs like sending an email or posting to social, is there an approval step before it acts? A per-TODO preview of the planned steps would build trust early.

Running the agent in your own browser, on accounts you already have (email, docs, code), is a different model from a hosted assistant that needs a new OAuth grant for every tool. The examples — reply to 3 open support tickets, ship a pricing-page fix — make the free tier versus $20 / $100 / $200 easier to judge for a small team. How does it behave when a TODO needs a confirmation, like actually sending an email, versus only drafting it?

Running the agent in your own browser, on accounts you already have (email, docs, code), is a different model from a hosted assistant that needs a new OAuth grant for every tool. The examples, reply to 3 open support tickets or ship a pricing page fix, make the free tier versus 20, 100 and 200 dollars easier to judge for a small team. How does it behave when a TODO needs a confirmation, like actually sending an email, versus only drafting it?

The part I find most interesting is that it starts proposing TODOs after learning from what I accept and reject. Can I restrict which accounts a single TODO may use, for example let it draft LinkedIn posts but never touch ad spend or send emails without my approval?

really great concept

Running in my own browser with the accounts I already use is a big plus for trust. Before the agent acts on something like sending an email, does it ask for confirmation, or can I set that per task type?

Love the concept of a TODO list that actually thinks and acts on your behalf. The gap between task management and task execution has always been frustrating. Does it integrate with popular project management tools like Linear or Notion? Being able to auto-execute simple tasks would be a game changer for founders juggling too many things.

This is wht I seperately do while building my products. If there is a queue like setup, when I have a thought on something I will add them to the queue, one by one it is given to the claude code. so I dont forget the idea.

This is an amazing App. This is very Useful

Running in the user's own browser and accounts is useful, but it makes permissions important. Can you require approval before high-impact actions such as sending emails or posting publicly? A short log of what the agent did and why, with an undo option where possible, would build trust quickly.

Running the agent in your own browser with the accounts you already use is a smart way to avoid yet another integration setup. How does it ask for confirmation before doing something irreversible like sending an email?

The idea of agents executing directly from a plain TODO list is compelling — it removes the whole prompt-engineering step from getting work done. Running in my own browser with accounts I already use is also a smart trust model. I'm curious how it handles tasks that need a judgment call mid-way: does it pause and ask, or make its best guess and let me correct it?

Using the accounts I'm already logged into, instead of new OAuth grants, is a nice touch. Two practical questions: what happens if a site asks for a 2FA code halfway through a task? And do scheduled TODOs (like the weekly LinkedIn update) still run if my laptop is closed, or does the machine need to be on?

Having the agent run in my own browser with my existing accounts is the part that sold me, since it skips the usual "connect 12 integrations" setup. How does it handle a TODO that's ambiguous? Does it ask before acting, or try and let you reject it?

Having the agent run in my own browser with my existing accounts is the part that sold me, since it skips the usual "connect 12 integrations" setup. How does it handle a TODO that's ambiguous? Does it ask before acting, or try and let you reject it?

The browser-local execution is the part that makes this credible for admin work: the agent uses accounts the person already trusts instead of asking them to hand over another cloud token. One thing I would want before delegating a recurring task is a plain audit view showing which account and which tool each run touched, plus the exact diff or sent message. Is that on the roadmap?

The idea of a TODO list that actually thinks and acts is an interesting shift from tools that just store tasks.

Running the agent in your own browser so it reuses accounts you already trust is a practical design for small teams. When it proposes TODOs on its own, can you set per-category approval rules, such as auto-run drafts but always ask before sending emails or posting? That would make it easier to trust on day one.