Gary Somerhalder (@garysomerhalder) header
Gary Somerhalder (@garysomerhalder) profile picture

Serial entrepreneur & investor | Founder & CEO of Legatus SI 🏛️ Build your software empire for the age of superintelligence. | COMMAND · GOVERN · VERIFY

Nashville, TN.
Joined February 2012
Photo by Gary Somerhalder (@garysomerhalder)Photo by Gary Somerhalder (@garysomerhalder)Photo by Gary Somerhalder (@garysomerhalder)Photo by Gary Somerhalder (@garysomerhalder)Photo by Gary Somerhalder (@garysomerhalder)Photo by Gary Somerhalder (@garysomerhalder)Photo by Gary Somerhalder (@garysomerhalder)
Here’s a cost I think deserves more attention in autonomous software development. How much work does it take to establish that an agent’s change is acceptable? Someone has to reconstruct the objective. Find the relevant evidence. Check what actually ran. Resolve the gaps. Decide whether the result can ship. If that effort grows with every agent we add, verification becomes a scaling constraint. That’s a problem I’m building Legatus to address. The investment thesis is infrastructure that helps enterprises scale verified delivery. If you invest in developer tools or cybersecurity, let’s talk. 🏛️
2
6
164
A cryptographic record can prove that a statement hasn’t changed. It doesn’t automatically prove the statement was correct. That distinction matters for AI agents. An immutable record of “tests passed” is only useful if we can establish which tests ran, against which revision, and whether they checked the requirement. I want the evidence connected to the actual execution and independently evaluated. That’s how I’m thinking about proof inside Legatus. Preserve the record. Verify what it establishes. Technical investors interested in agent governance: this is the architecture conversation I’d love to have.
2
88
I don’t want autonomous software development to mean a human approving every tool call. That would turn the human into the bottleneck. I want clear rules established before the work starts. Let agents execute within those rules. Escalate when they need more authority. Require independent evidence before consequential transitions. The purpose of governance is to make useful autonomy possible at scale. That’s the product direction I’m pursuing with Legatus. If you’re a pre-seed investor thinking about how enterprises move from agent experiments to autonomous engineering, I’d love to compare notes. 🏛️
1
1
44
Here’s the enterprise buying conversation I think autonomous agents will create. Engineering wants more work completed. Security wants enforceable limits. Leadership wants accountability when something goes wrong. All three need to understand the same run. What was the objective? Which actions were authorized? What changed? What independently verified the result? I believe the infrastructure connecting those answers will become increasingly valuable as agents take on more responsibility. That’s the opportunity I’m pursuing with Legatus. If you invest where developer infrastructure meets cybersecurity, I’d love to show you our approach. Command. Govern. Verify. 🏛️
30
A production incident creates pressure to move fast. An agent proposes an emergency fix. Someone approves an exception to the normal process. The immediate problem gets resolved. But when does that exception expire? Who verifies the temporary access was removed? Who checks that the security gate was restored? Emergency authority needs an end condition. That’s the kind of engineering detail that matters when autonomous agents operate real systems. At Legatus, I’m thinking about governance through the entire lifecycle—including what happens after the emergency is over. Investors building a thesis around agent security: my DMs are open.
1
23
An AI agent delegates a task to another agent. That agent delegates part of it again. Three agents later, someone is modifying production infrastructure. Where did that authority come from? Who approved the delegation? Did the original restrictions survive every handoff? Multi-agent systems need more than shared context. They need a chain of authority that stays enforceable as work moves between agents. That’s an important part of the opportunity I see with Legatus. The more autonomous the workforce becomes, the more valuable its control plane could become. 🏛️
3
1
27
An agent gets approval to deploy. Then another agent changes the code. The first agent still has the approval. Should it still be allowed to deploy? This is where agent governance gets interesting. Approval needs to belong to a specific change, under specific conditions, with specific evidence. When those conditions change, the approval needs to be evaluated again. That’s the kind of problem we’re tackling at Legatus. As engineering becomes more autonomous, authorization has to stay connected to what is actually happening. Cybersecurity and developer infrastructure investors: I’d love to compare theses.
5
An autonomous agent is stopped during a security incident. Credentials are revoked. The immediate threat is contained. Then someone restores access. What happens to the jobs that were already queued? Do they resume automatically? Do their approvals still apply? Has anyone checked whether the original objective is still valid? Stopping an agent is only half the problem. Restarting it needs its own authorization. That’s the level of lifecycle governance I’m building toward with Legatus. If you’re investing in the infrastructure that makes autonomous systems governable, let’s talk. 🏛️
4
An API key answers: can this agent access the system? It doesn’t answer: is this action authorized for this objective, against this target, with this evidence? That gap is where I’m building Legatus. Agent permissions need to become action-level decisions.
10
Gary Somerhalder retweeted
@garysomerhalder yeah agent governance is about to be a real budget line. going through my network for the funds that see it
1
1
41
A useful test for agent governance: can the agent grant itself more authority? If it can edit the policy, bypass the gate, and approve its own deployment, the boundary is incomplete. Legatus is built around a simple principle: authority must be enforced outside the agent.
10
I’m building Legatus for a future where AI agents do most of the software engineering. They’ll architect, code, test, review, and deploy. But who governs them? Who independently verifies their work? Who proves they followed the rules? We’re building the control plane for autonomous software development. Policy enforcement. Independent verification. Tamper-evident proof. I believe this will become foundational AI infrastructure. If you’re a technical VC investing in cybersecurity, developer tools, or autonomous AI, let’s talk. 🏛️ Command. Govern. Verify.
14
The AI industry is moving from copilots to autonomous operators. That’s a much bigger transition than most people realize. A copilot recommends. An autonomous agent acts. And when software starts acting on its own, security can no longer rely on recommendations and good intentions. It needs enforceable boundaries, independent verification, and trustworthy evidence. That’s why we’re building Legatus. The future of AI isn’t just more autonomy. It’s verifiable autonomy.
11
If you’re investing in autonomous coding agents, here’s a question worth asking: What happens when the agent does exactly what it’s capable of doing, but not what the organization intended? It modifies the wrong system. Weakens a security requirement. Skips a review. Deploys an unverified change. More intelligence helps. But enforceable governance and independent verification solve a different problem. That’s the infrastructure we’re building at Legatus. Would love to compare theses with investors in this space.
21
I’m making a very specific bet. AI agents will become capable enough to build and deploy increasingly complex software with minimal human intervention. But enterprises won’t surrender control of their engineering standards, security requirements, or production environments. They’ll demand infrastructure that makes autonomy governable. That’s what I’m building at Legatus. Not a replacement for AI agents. The control plane that makes their work verifiable. Command. Govern. Verify. 🏛️
1
20
Every autonomous coding platform needs trust. Every enterprise needs governance. Every security team needs evidence. But none of those requirements should depend entirely on one AI model or coding assistant. That’s why I believe independent agent governance is such a compelling infrastructure opportunity. Legatus is being built as a model-agnostic control plane for autonomous software development. One governance layer. Across agents, tools, and engineering workflows. Investors thinking about this market: let’s talk.
27
The security perimeter is changing. It used to be primarily about networks, identities, and applications. Now AI agents operate across all three. They read repositories. Call APIs. Execute commands. Access cloud infrastructure. Make changes without waiting for human approval. The new security question is whether every consequential action stays within policy. That’s the problem Legatus is being built to solve. The next perimeter is the agent’s authority.
21
An AI agent reports: “Security checks passed.” But what does that actually prove? Did the agent run the original tests? Did it modify the test configuration? Did it bypass a required review? Did it validate the production outcome? An agent’s claim of success is not independent evidence. We’re building Legatus to separate execution from verification. Because trustworthy autonomous software requires more than a green checkmark.
13
I’m building Legatus because I believe we’re witnessing the creation of an entirely new enterprise infrastructure category. Autonomous Agent Governance. AI agents will write software, coordinate engineering teams, manage infrastructure, and deploy applications. But enterprises will need independent control over what those agents are allowed to do. That’s where Legatus comes in. Policy enforcement. Verification. Cryptographic proof. I’m looking to connect with investors who see the same opportunity. If you’re building an investment thesis around autonomous AI, my DMs are open. 🏛️
14
Railway. PHP. Node.js. AI tooling. Look at how much critical infrastructure depends on software that can develop serious security flaws. Now imagine autonomous agents modifying and deploying that infrastructure continuously. The answer isn’t to stop using AI. It’s to build security governance into the autonomous development process. Enforce policies before actions. Verify outcomes independently. Preserve evidence. That’s the mission behind Legatus. Autonomy needs accountability.
31