Nobody built a monitor - Slack Code, the code channel and section 87 BetrVG

Salesforce shipped Slack Code on 20 August 2026: tag a coding agent in any Slack conversation and it opens a code channel holding the conversation, the plan, the code diffs and a live preview, at no extra cost on every Slack subscription. The governance answer on offer is an access-control answer - agents inherit Slack's permissions model and admins can gate who ships code. In Germany the first binding question is a different one. Section 87(1)(6) BetrVG gives the works council co-determination over technical devices 'dazu bestimmt' to monitor employee conduct or performance, and the word reads like intent only until you read the case law: in 1 ABR 16/23, decided 16 July 2024, the Bundesarbeitsgericht held a listen-in headset system co-determined 'auch dann, wenn die Gespraeche nicht aufgezeichnet oder gespeichert werden' - objective suitability decides, the employer's monitoring intention does not. A code channel is attributed by construction: who asked, what the agent returned, how many rounds, who approved. The AI Act's own workplace clause, Article 26(7), binds deployers of high-risk systems, and Regulation (EU) 2026/1744 moved Annex III to 2 December 2027. Section 87 has applied since 1972, with no risk tier and no transitional period.

Post

Salesforce shipped Slack Code on 20 August. It puts a coding agent, a code diff and a named human sign-off in the same scrollback — and in Germany that is a co-determination question before it is a security one.

Tag a coding agent in any Slack conversation and it opens a code channel: the conversation, the plan, the code diffs, a live preview. Claude Code, Devin, GitHub Copilot and Vercel at launch — Salesforce's own page also names ChatGPT. At no extra cost, on every Slack subscription.

Admins get guardrails, and agents inherit Slack's permissions model. Both true. Both answers to a security question.

Now the German part, and the correction.

§ 87(1)(6) BetrVG gives the works council co-determination over the "Einführung und Anwendung von technischen Einrichtungen, die dazu bestimmt sind, das Verhalten oder die Leistung der Arbeitnehmer zu überwachen." Read once, "bestimmt" sounds like intent: you did not build a monitor, so it does not apply.

That is not how it is read. In 1 ABR 16/23, decided 16 July 2024, the Bundesarbeitsgericht held that a headset system letting supervisors listen in is such a technische Einrichtung, and that it is co-determined "auch dann, wenn die Gespräche nicht aufgezeichnet oder gespeichert werden" — even where nothing is recorded or stored. Objective suitability decides. The employer's monitoring intent does not.

A code channel is attributed by construction. Who asked. What the agent returned. How many rounds it took. Who approved it. Searchable, per developer.

The AI Act does have a workplace clause — Article 26(7), inform workers' representatives before putting a high-risk system into service. It binds deployers of high-risk systems, and Regulation (EU) 2026/1744 moved Annex III to 2 December 2027. § 87 has been in force since 1972. No risk tier, no transitional period, and it attaches wherever a works council exists.

What I'd do Monday:

  1. Check whether your existing works agreement names anything like a code channel. "Slack" is not the same object.
  2. Default code channels off. Opt in per team, in writing.
  3. Write down what is retained, for how long, and who can search it. Before the first channel.
  4. Ask who is named as approver. That field is a performance record.

Nobody set out to build a monitor. Under § 87 that was never the test.

Whose sign-off is in your scrollback?

#AgenticAI #EUAIAct #Betriebsrat #SoftwareEngineering #Compliance

First comment

Sources.

Salesforce's own announcement — what a code channel contains, the founding agent partners: https://www.salesforce.com/introducing-slack-code/

Computerworld — session artifacts, admin guardrails, "no extra cost on all Slack subscriptions", and Forrester on permission inheritance: https://www.computerworld.com/article/4212446/new-slack-code-turns-ai-coding-into-a-team-activity.html

SiliconANGLE, 20 August — launch coverage and the canvas/diff flow: https://siliconangle.com/2026/08/20/salesforce-introduces-slack-code-to-bring-agentic-team-coding-into-the-open/

Bundesarbeitsgericht, 1 ABR 16/23, 16 July 2024 — the Leitsatz, in the court's own words: https://www.bundesarbeitsgericht.de/entscheidung/1-abr-16-23/

§ 87 BetrVG, full text: https://dejure.org/gesetze/BetrVG/87.html

AI Act Article 26(7), the workplace information duty: https://artificialintelligenceact.eu/article/26/

One point I cut for length: five vendors' agents can be tagged into channels across one workspace, and each is a separate processor relationship. Which one handles a given task is decided by whoever types the @-mention.

Claims

Sources

Corrections

None.

AI disclosure

The video for this edition contains AI-generated media: the narration is a synthetic voice and the cover art is generated. The script, the claims and the source verification are human-authored and human-checked. Assets carry IPTC DigitalSourceType and C2PA Content Credentials.