Layer 8 · Agents
Systems that pursue a goal over many steps: plan, act with tools, check the result, repeat.
In simple words: AIs that are given a goal and work through the steps themselves: plan, act, check, repeat.
Like: A new employee given a goal instead of a script: they work out the steps, use the office tools and check their own work.
Where it breaks: An employee builds judgement over months. An agent's 'experience' is only what fits in its context and memory, and it can repeat a mistake confidently.
Key players
What it means for your business: Agents can save real staff time but act inside your systems, so start with work that is easy to check, keep a person approving risky steps, and expect usage costs to vary.
- Claude Code: Anthropic's coding agent: reads a codebase, edits files and runs commands.
- OpenAI Codex: OpenAI's coding agent, as an open-source terminal tool and a cloud service.
- GitHub Copilot coding agent: Works on an assigned task in the background and opens a pull request.
- OpenHands: Open-source coding agent platform (MIT licence).
- OpenClaw: Open-source personal agent you host yourself, reached through chat apps.
- Salesforce Agentforce: Agents for customer service and sales, built on a company's CRM data.
What it gives the layer above
Completed multi-step tasks, not just single answers.
What it needs from the layer below
Tools, retrieval and model calls arranged by Orchestration & Tooling.
Concepts
- The agent loop
Plan, act with a tool, observe the result, and repeat until the goal is met. - Agent harnesses
The program wrapped around a model that turns it into an agent: it runs tools, asks permission, manages context and keeps memory. - Coding agents
Agents that work inside a code repository: read files, make edits, run commands and tests, and iterate until the checks pass. - Computer & browser use
Agents that operate a screen like a person: look at a screenshot, then click, type and scroll. - Multi-agent systems
Several agents with different roles or contexts work together on one task.
Walkthrough
- A goal, not a script. Someone asks: “fix the failing login test”. The agent loop starts: read the code, choose a step, act, look at the result, repeat.
- The harness keeps it safe. The harness runs each tool call, asks before risky commands, and trims the context so the long task still fits.
- A coding agent does the work. Inside the harness, a coding agent edits two files and reruns the tests until they pass.
- Checking the real page. To be sure, it opens the login page in a browser, takes screenshots and clicks through it, like a person would.
- Splitting the job. For a bigger change, a lead agent hands tests and review to sub-agents, each running its own loop.
Examples
- Agent harnesses (Claude Code, OpenCode, OpenClaw, Hermes Agent)
- Coding agents
- Computer and browser use
- Research agents
- Multi-agent systems