Mediapilote Backlinks, AI agent-driven link building

Link building involves a lot of manual work: auditing a site’s links, finding relevant outlets, contacting them, negotiating, checking publication, then making sure the link stays live. Mediapilote wanted to turn it into a product, with a client area and built-in invoicing.

The project started in late August 2026 from a product specification and a complete mockup. It relies on the intranet for accounts and on the internal SEO tool for backlink data.

The core of the product is a long, asynchronous process: an exchange with an outlet can stretch over several weeks, with follow-ups, unexpected replies, refusals or disputes. Every step had to survive restarts and hand control back to the team when a decision falls outside the agreed scope.

The agent writes to real outlets on the agency’s behalf. Its emails had to stay plain, be transparent about their purpose and honour opt-out requests, and the team had to be able to test everything without sending anything.

I set up a Cloudflare monorepo with a Next.js web app and an agent Worker, built in sprints: authentication, client portfolio, onboarding and weekly audits, source discovery and scoring, acquisition, article generation, client area, Stripe payments and Factur-X invoicing, link monitoring and disputes. Each process is a durable Workflow, split into atomic steps to stay within duration limits.

Outgoing emails go through automatic checks (length, personalisation, opt-out link), and a simulation mode redirects everything to a test address. Incoming replies are classified by AI and attached to the right opportunity. The interface was compared screen by screen with the mockup until every blocking gap was closed.