Gemini Cloud Assist (parked — revisit at second GCP project)
DARE.CO.UK · PARKED SKETCH · 2026-05-18
Mirrored from ~/.claude/.../memory/project_gemini_cloud_assist_parked.md. This is a design sketch parked for future build — read for context, not as a current deliverable.
GCP operational AI assistant (logs, IAM, cost, architecture); parked 2026-05-13 at one-project scale; resume signal = second GCP project lands or unexpected drift in current setup
Gemini Cloud Assist is GCP’s operational AI layer — natural-language interface to logs, IAM hygiene checks, cost optimization, architecture review against the state of your GCP environment. Distinct from Vertex AI Search (operates on document corpora); Cloud Assist operates on the state of your cloud.
Why parked 2026-05-13: at one-project scale (audrey-experiments only, devcorpus parked, dogwood and audrey production still on Cloudflare not GCP), there’s not enough cross-project drift, unused-resource volume, or log volume for Cloud Assist to find anything interesting. Value scales with footprint — Dan’s honest read 2026-05-13: “park it, revisit at second GCP project.”
Resume signal: second GCP project lands. Most likely triggers:
- Customer-facing AI voice work needs its own project (per feedback_customer_voice_over_internal_tools)
- A client engagement opens a client-<name> folder per the agency-tenant IAM model in reference_gcp_cli_and_agency_iam
- xlab-corpus (or similar) graduates out of audrey-experiments per project_xlab_co_lifecycle_model
Pricing at park time: paid tier ~$19-25/user/month; free tier covers basic console use. Verify pricing freshly when un-parking — GCP AI feature pricing moves quickly.
Counter-signal that would override the resume threshold: unexpected cost spike, IAM incident, or operational confusion in the current single-project setup that Cloud Assist would have caught. In that case, evaluate earlier than the formal resume signal.
Disposition pattern: feedback_park_with_resume_conditions. The mockup-equivalent here is the 2026-05-13 comparison-table chat (Vertex AI Search vs Cloud Assist vs voice-AI work) — captured in the devcorpus session, no separate artefact needed.