The stack, as of August 2026
The assistant
One AI coding assistant, run from the terminal against a git repository. It reads the project instruction file before it touches anything, and every job on this page runs through it.
The connection standard
The connections below use the open standard that has emerged for giving assistants access to other software. Where a system ships a connector, that connector is the connection. Where none exists, the connection is built here against the system’s API; four entries below are connected that way.
This site
Next.js, statically exported. A private GitHub repository. Cloudflare, deploying on every push to the main branch. The assistant edits the repository and ships by pushing; the reference entries went live this way.
Marketing data
An SEO platform (vendor connector). Ranking and link data are read straight from it during search work.
Web analytics (vendor connector). Traffic questions are answered from live data.
Google Ads, Meta Ads, TikTok Ads and Pinterest Ads (self-built). Spend and results for the monthly client reports are pulled from the platforms’ APIs, with no export step.
Client records and documents
A CRM (vendor connector). Pipeline and contact history are read when client work calls for them.
A document and spreadsheet store (vendor connector). The lead-quality sheet the monthly reports join against lives here.
The working day
Email and calendar (vendor connectors). Read for context when the assistant orders the queue.
Team chat (vendor connector). Read for context on running work.
An automation platform (vendor connector). The scheduled hand-offs between the systems above live here; the assistant reads and amends them.
Speech
Wispr Flow. Dictation all day; meetings become transcripts, and transcripts become actions and tasks in the project system below.
Own-built systems
A project management system, built here with the assistant, which has access to it. The assistant sets the order of the queue and groups tasks so a day can be spent as one block for a single client. Time tracking lives in it too: when a commercial time-tracking tool’s app stopped working, the assistant exported the history and tracking was rebuilt into this system.
Feed Engine Group’s affiliate tracking, built to replace a commercial tool so the group owns the data. The assistant works on it directly: new programmes are added through it, and trend readings and partner communication start from its tables.
Scheduled runs
Cloud-based routines that run without a machine at this desk being open. Work continues while I am away; I check in from my phone.
Why this page exists
This page holds everything in the section that dates fastest, which is its job; the other pages point here for current facts. Last reviewed: 31 August 2026.