The comparison

Frode Stenstrøm and Adrian Sjøholt gave Google Gemini, Microsoft Copilot and Claude the same task: contributing content to Norway’s standard cloud agreement. The demonstration compared workflow, capacity and the need for precise instructions rather than naming a universal winner.

What this means for the organisation

Contract work requires checks against the correct agreement version, the actual delivery conditions and legal review. AI can assist ideation, structure and first drafts, but the source of each requirement and responsibility for the final wording must remain clear.

The comparison gave Gemini, Copilot and Claude the same Lille Sky task, making differences in workflow and output visible in one session. The test concerned specific model versions and prompts at the recording date. It therefore says most about this task, while legal responsibility and the contractual basis still require separate review.

The Norwegian Digitalisation Agency’s SSA page is the authoritative entry point to the templates and guidance used in the test. It defines the contractual basis, while the session examines how three AI tools assist drafting. The distinction matters: a model may propose text, but the agreement version and actual obligations govern review.

An auditable AI test should retain the agreement version, prompt, model version and every manual change to the draft. A responsible specialist can then check requirements, pricing, liability and deviations against the actual delivery. Speed is only one measure; omissions discovered late can be much more costly in contract work.

Sources

Primary source: ProsessPilotene, “Få AI-hjelp med kontraktsarbeidet,” 30 September 2025 (Norwegian).

Norwegian Digitalisation Agency, Standard Agreement SSA-L, 2022 edition.

The original Norwegian transcript is retained as archival source material in the Norwegian version.

Join the discussion

What would you investigate first before using this? Share your questions or experience in the discussion.