Google Gemini API gets free tier and cost controls for agents

Building AI agents has always carried a hidden tax: the cost of getting it wrong. Unexpected API charges from a looping agent, complex hosting requirements, and the absence of safe sandboxes have made autonomous workflow development a risky experiment for smaller teams. On 28 July 2026, Google announced a set of updates to Managed Agents in its Gemini API that directly address those barriers. The headline additions are a free development tier, token budget controls, and scheduled triggers. For in-house tech teams at UK SMEs, the calculus on agentic AI just shifted.

What Google changed in the Gemini API

Google has made Gemini 3.6 Flash the default model for Managed Agents. That is the faster, leaner model in the Gemini family, and it now runs agent tasks by default rather than requiring developers to configure it manually.

Alongside that, three new capabilities arrived. First, a free tier gives developers access to Managed Agents at no cost during prototyping. Second, token budget controls let engineering managers set hard limits on how many tokens an agent can consume per run, preventing a misbehaving workflow from generating unexpected costs. Third, environment hooks allow teams to audit every tool call an agent makes before it executes, giving developers a structured way to inspect and validate agent behaviour in real time.

Scheduled triggers round out the update. Developers can now set agents to run on a recurring basis inside isolated sandboxes, removing the need for external cron jobs or third-party orchestration infrastructure.

Why this matters for UK engineering teams

For engineering managers at UK SMEs, the previous version of this decision involved a difficult trade-off. Prototyping agentic workflows meant either absorbing unpredictable API costs or waiting until a budget could be approved for a more controlled setup. Neither option encouraged rapid iteration.

The token budget controls change that directly. A looping agent, where a workflow calls a tool repeatedly without resolving, has been one of the most common causes of unexpected charges in agentic development. Hard budget limits mean a misconfigured agent hits a ceiling rather than an invoice.

The free tier removes the procurement conversation entirely for early-stage work. In-house tech teams can now build, test, and validate an agent workflow before any cost is incurred. That is a meaningful shift for smaller teams where a failed prototype would previously have required a post-mortem justification to finance.

Google's update also highlights that AI adoption in agent development is as much about confidence and culture as it is about capability. gecco's experience working with SME teams supports this: removing the financial risk of experimentation is often the step that allows technical staff to move from planning to building.

Three things your team can do this week

1. Run a test agent in the free tier. Set up a minimal Managed Agent task in the Gemini API using the free development tier. Aim for something low-stakes: a scheduled data validation check or a simple reporting trigger. This takes under 30 minutes with an existing Google Cloud account and gives your team direct experience of the sandbox environment before any costs are involved.

2. Set a token budget on every new agent build. Before deploying any agent to a live environment, configure a token budget limit at the project level. Treat this as a standard part of your internal build checklist, the same way you would set memory limits on a containerised service.

3. Use environment hooks to audit your tool calls. Map out the tools your agent will call and attach environment hooks to the ones that interact with external systems. This gives engineering managers a clear audit trail and a practical way to catch unexpected behaviour before it reaches production.

Considerations and limitations

The free tier is for development use. Production workloads will still incur standard Gemini API costs, so teams should model expected token consumption before moving any agent to live.

Managed Agents runs within Google's infrastructure. Teams with strict data residency requirements or those operating under sector-specific compliance frameworks should review Google's data processing terms before prototyping with production data.

The scheduled trigger feature runs inside isolated sandboxes. That is appropriate for many automation tasks, but teams building agents that require persistent state across runs will need to architect for that separately.

Your next step with agentic AI

If your engineering team is ready to prototype but uncertain about where autonomous workflows fit into your wider technical stack, that is exactly the kind of question the AI Readiness survey is built to surface. Taking the survey gives you access to 65+ free resources and a custom AI Readiness report. That report leads to a free 45-minute AI Readiness call to walk through your results in detail.

gecco's Training and consultancy programme works with in-house tech teams and engineering managers to build the knowledge and frameworks needed to move from prototype to production with confidence, without creating a dependency on external tools or vendors.

Website · LinkedIn · Case Studies · Newsletter