Service
AI agent production review
Formare AI's agent production review is a fixed-scope engagement for teams that already have a working Gemini, Vertex AI or Google ADK agent. We read the code and the deployment, test the failure modes that appear once real users arrive (session state, tool reliability, tracing, cost, security and EU data residency) and deliver a written plan that ranks the gaps and estimates the work to close them. It is quoted in writing within 48 hours of a free 30-minute call, and you can act on the plan with us, with your own team, or not at all.
Updated
Scope
Who the review is for
The review fits a team whose agent works in a demo or a pilot and who need to know what will break before real users, real data and real money are involved. It also fits a system already in production that has had incidents, runaway costs or no way to explain a wrong answer.
- A Gemini function-calling loop or an ADK multi-agent system that has not yet faced real traffic
- An agent in a pilot with a handful of users, where the next step is a wider release
- A live agent with unexplained failures, rising token spend or no tracing
- A system that must pass a security, data protection or procurement review before launch
What we check
Six areas, each with a test you can repeat
| Area | What we examine | Example of what we look for |
|---|---|---|
| State and sessions | Where session state lives, behaviour on restart, idempotency of tools with side effects, protection against concurrent claims | Sessions held in memory, so a Cloud Run restart ends every conversation |
| Tool reliability | Timeouts, retries, circuit breakers, error handling, iteration caps | A tool that times out silently, so the agent reports success on stale data |
| Observability | Logging of model and tool calls, tracing, an evaluation set, alerts | No way to tell which sub-agent produced a wrong answer |
| Cost | Token use per session, history growth, caching, concurrency and instance caps, budget alerts | The full conversation history re-sent on every turn |
| Security and access | Service accounts, secrets, authentication between services, tool-level authorisation, handling of untrusted model output | Authorisation enforced in the prompt instead of in the tool |
| EU data handling | Region locks, an inventory of stores that hold personal data, the erasure path including the vector index, disclosure that users talk to an AI | Personal data copied into a log and a cache that an erasure request never reaches |
The same checks are published as the AI agent production readiness checklist for Google Cloud, so your team can run them itself. The review adds the part a checklist cannot: someone who has seen these failures reads your code and your deployment and tells you which gaps matter most.
Deliverables
What you receive
- A written report that ranks every gap by risk, with the evidence for each
- For each gap, the fix and an effort estimate, so the work can be planned and priced
- A target architecture diagram for your agent on Google Cloud
- A run-cost estimate at the traffic you expect
- A 60-minute readout with your engineers
Inputs
What we need from you
- Read access to the code repository
- Access to the Google Cloud project, or read-only exports of the deployment configuration and logs
- A named technical contact who can answer questions during the review
- A sample of real or representative conversations, with personal data removed
The review does not change your code or your infrastructure. It is not a penetration test or a legal opinion: it covers the engineering controls that support security and data protection obligations.
After the review
Three ways to use the plan
- Close the gaps with Formare AI as a fixed-scope hardening project: sessions, tool reliability, tracing, cost controls, deployment and handover
- Close them with your own team, using the report as the work list
- Stop here: the review stands on its own
Once an agent is live, an engineering retainer covers monitoring, evaluation and cost tuning.
Cost and timeline
What the review costs
The review is quoted as a fixed-scope project: scope, duration and price are set in a written proposal within 48 hours of a free 30-minute discovery call. It is narrower than the 4-Week AI Strategy & Architecture Audit, which looks at a whole AI platform rather than one agent.
Pricing is €600–€900 per day, from €8,000 for a fixed-scope project paid in milestones, or from €4,500/mo for a retainer. Rates are in euros and VAT may apply.
Work
Agent builds, and how each is used
Client project
AI Airport Transfer Platform
George Travel — georgetravel.co.uk
Full-stack AI booking platform for a UK private driver. Gemini agentic loop with 11 custom tools handles geocoding, fare calculation, conflict-safe booking (PostgreSQL advisory locks), Stripe payments, Google Calendar, automated emails, real-time flight tracking, and weather. Interoperable via A2A protocol and MCP.
- 11 AI tools
- 0 double-bookings since launch
- A2A + MCP interop
Production figure from the live system, reported from launch to the date of the build article (15 May 2025). Conflicts are prevented by a PostgreSQL advisory lock around the availability check and the insert.
How it was built
Reference build
NL → BigQuery Assistant
Conversational data analysis — multi-agent
Ask BigQuery in plain English: an orchestrator routes to schema-aware NL→SQL and Pandas agents.
How it was built
FAQ
Questions about AI agent production readiness review
Cost, timelines and fit, answered before the first call.
What is an AI agent production review?
A fixed-scope engagement in which we examine a working agent against six areas (state, tool reliability, observability, cost, security and EU data handling) and deliver a written, ranked plan with effort estimates. The same checks are published in our AI agent production readiness checklist for Google Cloud.
Does the agent have to be built with Google ADK?
No. We review Gemini function-calling loops and ADK multi-agent systems alike, and most checks apply to any agent that calls tools. Our experience is on Google Cloud, so the review is most useful for agents that run, or will run, there; the discovery call covers whether it fits your stack.
Do you change our code during the review?
No. The review is read-only. Fixing the gaps is a separate, fixed-scope project if you want us to make the changes.
Is this a security audit or a legal compliance review?
No. It covers the engineering controls that support security and data protection obligations, such as access, secrets, region locks and erasure paths. It is not a penetration test or legal advice; your counsel should confirm how GDPR and the EU AI Act apply to your use case.
How much does an agent production review cost and how long does it take?
The review is quoted as a fixed-scope project in a written proposal within 48 hours of a free 30-minute call. Pricing is €600–€900 per day, from €8,000 for a fixed-scope project paid in milestones, or from €4,500/mo for a retainer. Rates are in euros and VAT may apply.
Read next