An agent that runs our AI watch every morning

AI changes every week: new models, new tools, announcements from the labs, feedback from practitioners. Following it seriously takes hours of videos and reading that a small team does not have. You end up reading what the algorithm pushes, and you learn about an important announcement in a meeting, from a client.

Every morning, it lists what has been published since the day before in a list of chosen sources: videos from the labs and from independent practitioners, blogs, new software releases, forum discussions, newsletters received in a dedicated mailbox. It summarizes each relevant video so nobody has to watch it, writes one sheet per notable topic and assembles the note of the day in the Drive.

Two short sections turn the note toward the outside: "For our clients", what the news changes for ongoing meetings and proposals, and a publication idea. The team receives a short email: the essentials in three paragraphs and the link to the note.

It runs on Appstrate, our open source platform: each run takes place in an isolated container, and the agent never sees the credentials of the services it uses.

Five rules, written down rather than left to intuition:

• The sources are a written list. Adding a channel or a newsletter is one line in a file. The agent does not go exploring the web on its own.
• Cover widely, filter nothing in silence. What is dropped is counted: we know what the note has seen, even what it does not report.
• A section exists only if the news carries it. On a day without a strong idea, the note writes "No idea strong enough today." rather than forcing a publication.
• It never publishes. The agent prepares the material. Whatever goes out under TRACTR's name goes through a person.
• A memory in a file. The agent notes what it has already read. A missed morning loses nothing, and nothing is summarized twice.

The email of an imagined morning (illustrative example, fictional content):

• In brief. A major vendor opens to all its users the connection of its assistant to company mailboxes. For our clients, this is the question that will come up in meetings: where does the data go, and who has access to it.
• Also. Two practitioners compare three automation tools on the same invoicing case. Common conclusion: the gap comes down to recovery after an error.
• Publication idea. What a small business should check before plugging an assistant into its mail.

• A useful watch is measured by what you do with it. The "For our clients" section pays off from the very next meeting.
• The right to say "nothing today" makes the rest credible. A note that always finds something important ends up not being read.
• The watch becomes a common good. The whole team reads the same note, and discussions start from the same ground.

The same pattern works for any watch that takes time you do not have: competitors, regulation, public tenders, supplier prices. A written list of sources, a daily summary, a section that says what it changes for you, and nothing goes out without a person.