It's time to stop thinking about better models and to start thinking about better systems to support them.
@Gartner_inc predicts that over 40% of agentic AI projects will be canceled by the end of 2027... not because the models got worse, but because the economics and the risk profile never closed.
This isn’t a value problem. It’s not even a pricing problem.
It’s a discipline problem.
There's too many agents. They're too expensive. And no one owns what they're doing.
You can’t control spend until you know how much you’re spending. And you can’t know what you’re spending if you don’t know what agents you’re accountable for in the first place.
@WSJ recently published a great article on agent sprawl (link in the comments).
In the article, the author highlights some of the scaling challenges we’ve been hearing from enterprise teams—
- Runaway token costs
- Lack of permissions
- Shadow AI 🌑
The headline frames this as a proliferation problem. And that's not wrong! But it's not the whole story either.
Too many enterprise leaders are focused on building agents. That's true. But what the subtext of this WSJ article makes abundantly clear is that, in 2026, we need to be far more focused on operating them.
We saw the same tension with containers before @kubernetesio.
The problem isn't that companies have too many agents; it’s that they don’t have a centralized control plane to manage and govern them at scale.
That's the problem we're determined to solve.
https://t.co/BDYChmIChP
How you build your agents has a lot to say about how you'll govern them as well.
In his latest, Guild engineer @text2goat will walk you through how he builds an LLM agent on Guild's control plane.
@Microsoft just turned off Claude Code because the usage costs were too high. @ServiceNow blew through their entire 2026 token budget before Memorial Day.
@Uber too.
@Gartner_inc now predicts that over 40% of agentic AI projects will be canceled by the end of 2027, not because the models got worse, but because the economics and the risk profile never closed.
This isn’t a value problem. It’s not even a pricing problem.
It’s a discipline problem.
You can’t control spend until you know how much you’re spending. And you can’t know what you’re spending if you don’t know what agents you’re accountable for in the first place.
- Audit trails.
- Tracked permissions.
- Clear and enforceable ownership standards.
- And the insights to respond quickly when your agents fail.
This is the neutral governance layer that agents have been operating without since GPT 3.
We decided it’s time to build it.
"When I was leading the developer infrastructure team at Meta, we were supporting tens of thousands of engineers, and we started to see the shift—from tools that assist to systems that actually take action... And once that starts to happen, the question changes pretty quickly." - @jevering
→ Ownership.
→ Access controls.
→ Visibility.
→ Cost management.
→ Audit-ability.
When it comes to production agents, the question isn't "does it work?"
The real question is, "who's controlling it?"
Building agents is the easy part.
Managing them safely at enterprise scale isn't.
That's the problem we're solving at Guild.
So, what does that really mean? We'll let James and founding engineer Vincent Durmont tell you in their own words.
https://t.co/NxpwEScN1K
In a world full of surprises, sometimes you just need a little determinism in your life.
In this quick demo, @text2goat will show you how to create deterministic type-script agents with Guild—and satisfy your maniacal need for control.
When agent outputs get wonky, the temptation is to go back to the prompt. But you can't fix non-deterministic problems with non-deterministic solutions.
The problem isn't a gap in intelligence. It's a gap in infrastructure.
Here are the four pillars of secure agent infrastructure—
1. Input constraints: if you can't validate what goes into the agent, you can't control what comes out of it either.
2. Permission scoping: agents are an extensions of their users; and so are their permissions.
3. Cost boundaries: a bill with no owner has no limits.
4. 360 observability: what does good mean for this agent? did it get there? why or why not? When an output is non-deterministic, you don't just need to know what it was supposed to do... you need to know what it actually did, why, and how it compares to other relevant responses.
Check out the link below to learn more, and find out how we're bringing determinism to a non-deterministic world with Guild.
https://t.co/7WEJGxDb5M
“Over decades working in infrastructure, I’ve seen this pattern repeat.
A new capability shows up, teams start using it everywhere, and before long they need systems to coordinate and control it… I think agents are heading in the same direction.” -@jevering
Right now the ecosystem is focused on frameworks and SDKs: LangChain, CrewAI, AutoGen, OpenAI’s SDKs, and dozens more.
But once they’re touching source code, production systems, internal tools, customer data, and financial systems, you start needing things like:
- identity and permissions
- auditability and tracing
- operational visibility
- cost and resource control
- safe rollout and rollback
The ecosystem is still focused on building agents.
Eventually it will focus on operating them.
That’s the problem we’re solving at Guild.
“Over decades working in infrastructure, I’ve seen this pattern repeat.
A new capability shows up, teams start using it everywhere, and before long they need systems to coordinate and control it… I think agents are heading in the same direction.” -@jevering
Right now the ecosystem is focused on frameworks and SDKs: LangChain, CrewAI, AutoGen, OpenAI’s SDKs, and dozens more.
But once they’re touching source code, production systems, internal tools, customer data, and financial systems, you start needing things like:
- identity and permissions
- auditability and tracing
- operational visibility
- cost and resource control
- safe rollout and rollback
The ecosystem is still focused on building agents.
Eventually it will focus on operating them.
That’s the problem we’re solving at Guild.
“Over decades working in infrastructure, I’ve seen this pattern repeat.
A new capability shows up, teams start using it everywhere, and before long they need systems to coordinate and control it… I think agents are heading in the same direction.” -@jevering
Right now the ecosystem is focused on frameworks and SDKs: LangChain, CrewAI, AutoGen, OpenAI’s SDKs, and dozens more.
But once they’re touching source code, production systems, internal tools, customer data, and financial systems, you start needing things like:
- identity and permissions
- auditability and tracing
- operational visibility
- cost and resource control
- safe rollout and rollback
The ecosystem is still focused on building agents.
Eventually it will focus on operating them.
That’s the problem we’re solving at Guild.
“Over decades working in infrastructure, I’ve seen this pattern repeat.
A new capability shows up, teams start using it everywhere, and before long they need systems to coordinate and control it… I think agents are heading in the same direction.” -@jevering
Right now the ecosystem is focused on frameworks and SDKs: LangChain, CrewAI, AutoGen, OpenAI’s SDKs, and dozens more.
But once they’re touching source code, production systems, internal tools, customer data, and financial systems, you start needing things like:
- identity and permissions
- auditability and tracing
- operational visibility
- cost and resource control
- safe rollout and rollback
The ecosystem is still focused on building agents.
Eventually it will focus on operating them.
That’s the problem we’re solving at Guild.
“Over decades working in infrastructure, I’ve seen this pattern repeat.
A new capability shows up, teams start using it everywhere, and before long they need systems to coordinate and control it… I think agents are heading in the same direction.” -@jevering
Right now the ecosystem is focused on frameworks and SDKs: LangChain, CrewAI, AutoGen, OpenAI’s SDKs, and dozens more.
But once they’re touching source code, production systems, internal tools, customer data, and financial systems, you start needing things like:
- identity and permissions
- auditability and tracing
- operational visibility
- cost and resource control
- safe rollout and rollback
The ecosystem is still focused on building agents.
Eventually it will focus on operating them.
That’s the problem we’re solving at Guild.