Is your task list a jumble of bug reports, follow-ups and half-formed ideas scattered across five apps? I build Notestream inside Notestream, so every one of those lands in a single queue, sorted the moment it arrives.
This post is a look behind the curtain. I want to show the real mechanics of how a one-person product gets maintained with its own tool, including the parts that are still rough.
Nothing here is a demo. These are the same doors, the same sorting and the same dispatch flow you get when you sign up, pointed at my own backlog.
Where my tasks come from
Most of what lands in my queue was never planned. It shows up while I am doing something else, and the only question is whether it gets written down before I forget it.
Some of it comes from me. A thought about the pricing page arrives mid-walk, or I notice a typo while reading an old post on my phone. Some of it comes from automated checks I run on a schedule, like a nightly check that flags a broken image on a blog post or a watchdog that notices a scheduled job did not run when it should have.
And some of it comes from other people. A reader emails to say a page is confusing, and that message becomes a task the same way a client email would for anyone else using the product.
Three doors, one queue
Every one of those sources reaches the queue through one of three doors. Which door depends on where I am when the thought shows up.
The AI assistant field is the one I use most. When I am at my desk, I type one sentence, hit submit and go straight back to whatever I was doing. There is no project to pick and no tag to remember.
Forwarded email handles everything that arrives in my inbox. A support message or a renewal notice from a vendor gets forwarded to my Notestream address, and the commitment inside it leaves my inbox for good.
The MCP API covers the automated sources. My scheduled checks send their findings straight into the queue through MCP, so a broken image becomes a task without me copying anything. When I am mid-build in a tool that speaks MCP, I can drop a capture in from there too.
Small captures, sorted on arrival
Each capture is small on purpose. It is one idea, one commitment or one deadline, written the way I would say it out loud.
"Fix the broken cover image on the deadline post" is a capture. So are "reply to the reader who asked about forwarding" and "renew the domain before the 20th." A long document with six thoughts tangled together is not, and I split those up before they go in.
The sorting happens the moment I hit submit. Notestream reads the sentence and files it as an idea, a commitment or a deadline right then, automatically, with no overnight batch job behind it. If there is a date in the sentence, the capture carries that date from its first second.
That is the part I would miss most if it went away. I never sit down for a weekly sorting session, because there is nothing left to sort.
How I assign tasks to AI from the queue
Once a capture is sorted, the next question is who does the work. For a lot of my queue, the answer is an AI tool, and dispatch is how it gets there.
I pick the capture and send it to the tool that fits. Claude Code takes most code fixes, and Lovable takes front-end changes to the site. Claude Cowork handles work that runs against my own files. The in-app LLM chat is where I send quick research or a first draft, using ChatGPT, Gemini or Claude depending on the job.
The capture travels with its original sentence, so I never rewrite the context as a prompt. Notestream handles the classification on its own. Which AI finishes the work is my call, task by task.
The review step that keeps me honest
Dispatched work does not close itself. When a tool finishes, the task comes back with a comment describing what it found and what it changed, and it waits in a ready-for-review state.
I look at the result before I mark anything complete. Some fixes are clean on the first pass. Others need another round, and the comment trail on the task shows me why without any digging.
Some tasks stall entirely. They need a decision only I can make, or they touch something risky enough that an automated pass should not guess. Those stay open with a note explaining what is missing, and I would much rather see an honest open task than one marked done by mistake.
What a normal day looks like
On an ordinary day, the queue is not dramatic. A handful of tasks move from pending to ready for review, I confirm and close a few, and one or two sit open waiting on me.
The morning routine is short because the sorting is already done. I read the list once, handle anything that takes two minutes, schedule what needs real time, and dispatch the rest before I start on work only I can do.
That last part is where the hours come back. Work I hand off at eight in the morning is usually waiting for review by the time I come up for air.
What building in public taught me about the product
Running my own backlog through Notestream shows me the rough edges before anyone else sees them. When a capture lands with a label I did not expect, I notice right away, because it is my own list that gets messy.
It also keeps me clear about what the product is for. Notestream is a place where small captures land, get sorted on arrival and go out to whichever AI should do the work. That narrow job is the whole point, and looking at my own queue every day keeps me from promising more than that.
One task, start to finish
Here is a typical example of what moves through the queue. Overnight, a scheduled check sends in a capture through the MCP API: "Cover image on the deadline post returns a broken link." It lands as a commitment near the top of my list.
In the morning I dispatch it to Claude Code with the original sentence attached. Claude Code traces the broken link to a bad image path, fixes it and leaves a comment explaining the change. The task then waits in ready for review.
I open the post, see the image loading, and mark the task complete. My part takes a couple of minutes, and most of that is reading the comment.
If your own task list is spread across too many places, try running it through one queue for a week. Try it free at notestream.ai.