The draft is solved. The operation is not.

Language models settled one thing that agencies used to bill for: the first usable draft. Writing a post about an employment-law ruling now takes seconds. That is exactly why the draft is no longer the interesting question. The interesting question is what comes after, and that is where most attempts fail.

Between the draft and a published, approved, multi-channel post lies the real work: proposing topics that fit the firm, organising the approval, publishing at the right time, catching errors when a platform API misbehaves, following up when a slot stays empty. Someone has to build that route and then keep it alive, week after week.

One operated route, a dozen lawyers, two months.

Since mid-May 2026 we operate the social-media content production of a mid-sized law firm in the German-speaking region, around a dozen lawyers. Before, posts were made one by one, text in a chat tool, graphics by hand, everything hanging on individuals. Today an operated system proposes topics and finished drafts straight into Microsoft Teams, each lawyer approves their own posts with one click, publishing runs automatically. Operation, monitoring and maintenance sit with us.

In the two months since 21 May: 17 posts approved and published, in nearly 40 successful publications across Facebook, Instagram and LinkedIn. Around ten lawyers in the rotation, nine with their own published posts. Reach is honestly small, roughly 2,600 people reached, some 300 reactions. The point is not the reach, but the route that produces it reliably and without manual work.

Five lessons that transfer.

For anyone who needs content regularly, whether a firm, a practice or a mid-sized business.

01

The draft is solved, the operation is not

A model that produces text is a tool, not a solution. Whoever automates only the draft and does the rest by hand has gained nothing.

Operation
02

Rejection is the feature, not the bug

31 rejected against 17 published: the lawyers reject more than they approve. That is the proof that humans curate and no machine posts blindly. A system that cannot be rejected is useless to a firm.

Curation
03

Approval must go where people already are

Approval happens in the Microsoft Teams chat they already keep open, not in yet another portal with its own login. Every extra login halves participation.

Adoption
04

No central bottleneck

Each lawyer approves their own posts, not a central desk for all. That is why it scales to ten people instead of clogging at one. A failure in one place does not stall the whole route.

Scaling
05

Measure, do not claim, even when the numbers are small

The system records reach and reactions per post and per channel. That keeps the early numbers honest and improvable, instead of inflating them.

Measurement

The value is in the boring continuous operation.

This is not a virality story. The reach is that of a regional firm, not a publisher. The value of such automation is not in the clever prompt but in the boring continuous operation: a route that runs for two months without starting from zero every week, carried by ten people and monitored in the background.

A draft is a moment. A route that runs for two months, carried by ten people, is an infrastructure.

Every number verifiable.

  • All figures are pulled live from the running system (as of 23 July 2026) and backed with evidence on request.
  • The firm stays anonymous because many practices do not, right now, want to advertise publicly that they outsource their content production. That discretion is part of the work.
  • The same discipline we apply to a client's confidentiality is the one we would apply to yours.