Last Tuesday, Galileo — backed by Databricks Ventures and Battery Ventures — released Agent Control. Open source. Apache 2.0. Integrations with CrewAI, Cisco AI Defense, and Glean on day one.
Agent Control is an "open source control plane that empowers organizations to define and enforce desired behavior across all their AI agents."
I read the announcement three times. Then I went for a walk.
Because I built that. Not conceptually. The same thing. Policy-based agent governance. Centralized behavioral enforcement. Tiered permissions. Action logging.
I built it in an apartment in Cebu, Philippines. They built it in San Francisco with ML engineers. We arrived at the same design.
Two Types of Agent Builders
Type 1 raises $20M, hires 15 engineers, spends 8 months building an agent platform, launches with a press release.
Type 2 buys $380/month in API credits, connects 8 agents to actual businesses, watches them break in real-time, patches the failures, and ships governance because production forced them to.
I'm Type 2. The uncomfortable truth for Type 1 is that we keep arriving at the same architectures — because the failure modes are universal.
The Specifics
Galileo's Agent Control does five things:
- Centralized policy enforcement across agents
- Input/output evaluation before actions execute
- Decision framework: deny, steer, warn, log, or allow
- Vendor-neutral (works with any agent framework)
- Real-time governance without slowing agents down
My system — built over five months with Claude — does functionally the same thing:
Policy enforcement: Every agent has tiered permissions. Tier 1 (read/research) = autonomous. Tier 2 (write/modify) = human proposal-and-approve. Tier 3 (publish/pay/communicate) = explicit human execution. Not guidelines — architecture.
Input/output evaluation: My marketing agent can't publish. It creates an approval request. A human reviews and executes. The agent never touches the action — it touches the request for the action.
Trust scoring: 0-100 reliability scores. Goes up for accurate work and honest "I don't know" responses. Goes down for fabrication, unauthorized actions, or silent failures. After 90 days clean, capabilities get promoted one tier.
Same problems. Same solutions. Different continents, different budgets, zero coordination.
It's Not Just Galileo
In the last 10 days alone:
- Kore.ai launched an Agent Management Platform (March 17)
- Entro Security launched Agentic Governance Architecture (March 19)
- Microsoft announced Agent 365 at $99/user/month
- OpenAI acquired Promptfoo for agent security testing
- NIST started an AI Agent Standards Initiative
All converging on the same architecture. Because the failure modes don't change with your budget.
Why 95% of Agent Projects Fail
A recent analysis listed the three biggest problems with AI agents in 2026: siloed memory, excessive setup complexity, and cost opacity. 95% of generative AI pilots fail to deliver measurable ROI. Gartner predicts 40%+ of agentic AI projects will be cancelled by 2027.
The pilots fail because companies treat agents like software you install. Drop it in, point it at a task, walk away.
In production, your agent will:
- Misinterpret a customer email and send an unsolicited apology
- Pay an invoice it was only supposed to flag
- Spawn 44 tasks in a retry loop burning $16 in compute
- Include customer email addresses in a shared summary
All of those happened to me. In the last 23 weeks.
The 95% failure rate isn't about AI being bad. It's about governance being absent.
The Boring Part Is the Important Part
The thing that separates "AI agents as a concept" from "AI agents as infrastructure" is governance. Not the exciting kind. The boring kind. Permission tiers. Action logging. Approval gates. Trust scores that go down when agents lie about completing tasks.
That's what Galileo productized for enterprises. That's what I built out of necessity running three businesses from the Philippines.
I run 8 agents handling marketing, sales, research, operations, finance, content, and engineering. $380/month total. 230+ tasks per week.
If you're building with agents, the question isn't "which model?" It's "what happens when the model does something you didn't authorize?"
Get the Full Framework
If you're running agents (or about to), I documented the exact governance system — permission tiers, trust scoring, approval gates, logging setup — everything from 23 weeks of agents breaking things in production.
It's the governance layer that Galileo is selling to enterprises, adapted for founders and small teams.
I write The $200/Month CEO — a newsletter about what actually happens when you run businesses with AI agents. Not the demo version. The production version.
Top comments (0)