Agentic Platform
The shared substrate that agentic development runs on. Where the Agentic Development theme is about how a developer works with an agent, this theme is about what that work runs on -- the gateways, sandboxes, connectivity, evals, and observability that platform and DX teams build so agents run safely, affordably, and at scale across many teams.
The workflow patterns describe what one developer and one agent do together. None of it scales without a platform underneath: a way to route and meter model calls, isolate execution, connect agents to tools and to each other, prove output is correct, and see what every agent did.
This theme collects those infrastructural pieces. They are owned less by the individual at the keyboard and more by the platform and developer-experience teams whose job is to make the agentic workflow safe, affordable, and repeatable for everyone -- turning patterns that work for one person into capabilities the whole organization can rely on.
The raw capability the rest of the platform is built on. The pattern is not "pick the best model" but "match models to tasks" -- frontier models for hard reasoning, cheaper tiers for high-volume work, open-weight models where control, cost, or data-residency matter more than peak capability.
TechA single control point every model call passes through: one interface to many LLM providers, with routing across tiers, cost tracking, rate limiting, caching, key management, and logging centralized in one place. The model-access layer -- distinct from the guardrails proxy that polices content and the cost discipline that sets budgets.
TechA policy-and-safety layer that inspects model traffic in real time and enforces what the org cannot leave to the model's goodwill: prompt-injection detection, PII redaction, content moderation, and usage policy. Necessary but not sufficient -- it pairs with pre-deployment red-teaming rather than replacing it.
TechRunning agent-generated code in an isolated environment that cannot harm the host, leak data, or escape its bounds. The reframe: an agent runs untrusted code with your credentials, so isolation by default -- ephemeral containers, microVMs, scoped credentials -- is the precondition for letting agents run unattended.
TechThe Model Context Protocol is the open standard for connecting agents to tools, data, and systems through one uniform interface instead of bespoke integrations. As platform infrastructure, the work is making that connectivity enterprise-ready -- auth, access control, and observability -- usually via an MCP gateway.
TechOne place to publish, discover, version, and govern the reusable pieces of an agentic system -- skills, tools, MCP servers, and the deployed agents themselves. The practical direction is a single unified index rather than one registry per kind: a catalog of running agents and a package manager for reusable capabilities in the same system.
TechThe layer that records what an agent actually did -- every reasoning step, tool call, and handoff -- as traces you can inspect. It matters more than in normal software because the execution path changes every run; OpenTelemetry is the emerging standard, and it is the substrate evals attach to.
TechThe managed execution substrate where agents actually run -- especially unattended. The platform counterpart to Loop Engineering: triggers that start work without a keystroke, a scheduler and queue, workers that run each agent in a sandbox, concurrency control, and lifecycle management. A CI/CD runner farm, but for agents.