I built a marketing platform artificial intelligence runs
Giant Context is a marketing platform where the intelligence is the operator. A model reads what a business is, works out what its marketing is missing, and makes it. Seven apps on one dataset, nine months, one person.
This blog is the proof. Sixty-one posts and the page they sit on, all of it produced and published through the platform's own API.
What it does
Mind is the part of the platform that decides what work is needed. It reads a business the way an agency would if an agency had time to read everything, and it reads all of it every time it runs.
| Surface | What the platform does |
|---|---|
| Website | Builds it from the business's own files, then edits it when those files change |
| Blog | Writes the posts from the same material |
| Knowledge base | Writes the help articles, then finds the ones that went stale |
| Decides who gets what, one recipient at a time | |
| Chat | Answers visitors out of published material only |
| CRM | Keeps a contact history written for a model to read |
| Developer docs | Reads the real SDK source and writes the documentation from it |
Everything it makes arrives finished. Not an outline and not a draft to rewrite, but the page as it will appear, waiting on approval.
The website is built from the business's own uploaded files. Every concrete claim the model proposes has to anchor to that material above a similarity threshold, and anything that does not is dropped before it reaches the page. A model handed nothing but a request invents a testimonial no customer gave.
The email has no campaigns and no segments. Each email carries one sentence saying when it should go, and Mind reads a contact's own activity and decides. One person gets one ordered stream instead of four campaigns competing for the same inbox on the same morning.
The chat on a customer's site answers out of their published material only, carries the sources it drew on, and tells a visitor plainly when the material does not cover the question. It knows the difference between a customer's pricing and a competitor's, so it will never put the wrong number in front of a buyer.
The same pattern runs through the rest. The help centre is written and then checked back against what the business looks like now, the contact history is kept in a form a model can read, the developer documentation is written by reading the real SDK source, and every page updates when the files under it change.
Nothing goes out without approval
The platform does not publish on its own, and that is a decision rather than a limit. Autonomy is built and tested. The approval gate stays because a business handing over its public voice wants to see what goes out under its name, at least until it stops wanting to.
Two ways to run it
In the first, the platform decides. Mind reads the business, finds what is missing, produces the work, and the work arrives for approval.
In the second, a person directs it in conversation and the platform carries out the job over MCP. MCP is the protocol that lets a model call a service's functions directly. Giant Context generates its MCP server from the same OpenAPI description that generates the developer SDK and the console's own client, so an agent driving the platform uses the surface a customer's own integration uses. There is no separate AI mode, and no writer's API sitting beside the real one. An agent creating a post calls what a developer calls.
One specification describes 505 operations, and everything that touches the platform reads from it. Adding a capability to the API adds it to the SDK and to the agent's toolbox in the same deploy.
Driving it by hand taught me more about the autonomous loop than any test had. A job run under direction shows the order it actually wants to run in, and what came out of that goes into the autonomous mode over the next ten days.
The proof
This blog is what the directed mode produced. Nine months of commits, releases and abandoned drafts went in. What came out was a corpus with an argument, sorted into three acts.
| Act | What it answers | Posts |
|---|---|---|
| What I built, and how | The method, from the specification to the checks that let an agent deploy | 4 |
| What it looks like realized | The product a customer gets | 4 |
| What it looks like audited | What broke, what it cost, and what changed | 4 |
Every post was researched against real sources, drafted, and read back in full before it went live. Eight were cut, three of them for a claim at the centre that did not survive checking against the record. Nothing here published without me saying so, which is the gate a customer gets.
Every company has an unwritten blog
A company with nine months of shipped work has a blog it has never written. The material is already there, in commits, in release notes, in the documents on its own drive. What it does not have is the reading, the deciding and the revision that turn that into sixty posts a stranger would finish.
A blog is one surface of seven. The same machine that read nine months of commits and produced this corpus reads a business and produces its website, its help centre, its email and its answers to visitors. One person approving, and a marketing department's output behind it.
Marketing that runs itself. Software that heals itself, which already files my production errors with the fix attached. Software that builds itself. A platform that operates itself. A company that needs no humans at all.
Building something like this
I'm Jesse. I build platforms end to end, and I'm open to work. If this is the kind of engineering you need, get in touch.