# Notes on running AI agents across a company.

> Practical notes on running AI across a company: AI agents, gateways, guardrails, budgets and governance.

- Canonical: https://secondstack.ai/blog/

---

How companies give people and teams AI agents and keep models, budgets and records under control. Written by the team that builds SecondStack.

- [What is an AI agent?](https://secondstack.ai/blog/what-is-an-ai-agent/) (2026-11-06): An AI agent pursues a goal, plans its steps and uses tools on its own. A plain definition, how it differs from a chatbot or a workflow, and examples.
- [LLM governance: a working framework](https://secondstack.ai/blog/llm-governance/) (2026-09-09): LLM governance as five enforced layers: access, data boundaries, spend, visibility, change discipline, and the point in the request path where each control actually runs.
- [LLM cost management: budgets, alerts, per-team visibility](https://secondstack.ai/blog/llm-cost-management/) (2026-09-08): How LLM cost management works in practice: per-key attribution, budgets enforced in the request path, alerts before limits, per-team analytics.
- [LLM guardrails: PII and secrets before egress](https://secondstack.ai/blog/llm-guardrails/) (2026-08-26): LLM guardrails scan prompts for PII and secrets before egress. Block, mask, or log; where the policy attaches; and who in your organization owns each choice.
- [What is an MCP gateway?](https://secondstack.ai/blog/what-is-an-mcp-gateway/) (2026-08-24): An MCP gateway is a control point between AI agents and MCP servers: approved-tool catalog, per-team access control, credential brokering, audit log.
- [Claude Enterprise pricing: what you actually pay](https://secondstack.ai/blog/claude-enterprise-pricing/) (2026-08-12): Claude Team costs $20–25 per standard seat and caps at 150 seats. Enterprise starts at $20 per seat with a 20-seat minimum, plus usage at API rates. Full math, checked November 2026.
- [Shadow AI: what it is and how to detect it](https://secondstack.ai/blog/shadow-ai-detection/) (2026-08-10): Shadow AI is AI use without IT oversight. What it looks like in practice, what it costs when it goes wrong, and six practical detection signals.
- [Claude Code for enterprise: pricing, plans and governance](https://secondstack.ai/blog/claude-code-for-enterprise/) (2026-07-24): How Claude Code is licensed (seats vs API billing), what it costs per developer, the governance controls Anthropic ships, and where a gateway fits.
- [The company brain: own the history it will be built from](https://secondstack.ai/blog/company-brain/) (2026-07-23): What a company brain is, why tribal knowledge is the bottleneck, and why your LLM interaction history is the asset it will be built from.
- [ChatGPT Enterprise pricing: the full breakdown](https://secondstack.ai/blog/chatgpt-enterprise-pricing/) (2026-07-23): OpenAI doesn't publish ChatGPT Enterprise pricing. Reported 2026 deals: $50–60 per seat, 150-seat minimum, annual term. The math to run before signing.
- [What is an LLM gateway?](https://secondstack.ai/blog/what-is-an-llm-gateway/) (2026-07-14): Direct calls work — until a second team, provider, or coding agents arrive. What an LLM gateway does, what breaks without one, and when to add it.
- [Why we built SecondStack](https://secondstack.ai/blog/why-we-built-secondstack/) (2026-07-08): SaaS assistants store your data and bill per seat; DIY open-source stacks go unsupported. Why we built a self-hosted AI platform and ran it on ourselves first.