chapter five

5 From autocomplete to agent fleets: the road to Continuous AI

 

This chapter covers

  • Progressing from autocomplete to agents
  • Managing approval fatigue effectively
  • Running agents outside the IDE
  • Building team standards with AGENTS.md
  • Exploring agent orchestration tradeoffs

An AI agent is only as autonomous as the leash you're willing to give it.

We already explored context engineering — how to feed models the right information at the right time. But context is just fuel. The question remains, what kind of engine are we building?

To answer that question, we'll move from passive autocomplete, through agents that can see and touch your codebase, to headless automation in CI pipelines, and finally toward orchestrated fleets of specialized agents. At each step, AI moves further away from a coding aid for one developer and deeper into the SDLC (Software Development Lifecycle) itself.

Read that as an expanding menu, not a ladder you climb and kick away. None of these forms retired the one before it. Tab completion is still the most reliable option when you know exactly what you want to type, approval prompts did not vanish when agents moved to the terminal, and a five-agent swarm beats one agent only on a narrow class of parallelizable work. Which shape fits depends on the task, not on how recently the shape was invented.

5.1 Passive AI: ChatGPT as a coding partner

5.1.1 GitHub Copilot: AI moves into the IDE

5.1.2 Tab-Tab-Tab: the illusion of productivity

5.1.3 The context blindness wall

5.2 Agents in IDE, but still on leash

5.2.1 The Cambrian explosion of AI coding agents

5.2.2 Approval fatigue

5.2.3 Trade-off: vision vs. vigilance

5.2.4 Slopsquatting and package hallucinations

5.3 Headless era: Moving into terminal

5.3.1 Codex, Antigravity and Claude Code: AI in the UNIX pipeline

5.3.2 Agent as gatekeeper and reviewer in Git

5.3.3 Debugging a lost agent

5.3.4 Trade-off: autonomy vs. observability

5.4 Standardizing team workflows with agents

5.4.1 From .cursorrules to AGENTS.md

5.4.2 Agent handoff: passing context between agents

5.5 Governance and compliance

5.5.1 Agent policies: the missing standard

5.5.2 Governance needs an evidence layer

5.5.3 Trade-off: scalability vs. bureaucracy

5.6 Swarm of agents and orchestration

5.6.1 Poor man's parallelism: Git worktrees

5.6.2 Observability becomes critical

5.6.3 IDE support: parallelism gets easier (and harder)

5.6.4 Mission Control: Agent HQ and Antigravity

5.6.5 Artifacts as the currency of trust