Phase 1 of 4
Building the AI habit
Augment is installed but agent features haven't yet become part of the daily workflow. The team is in discovery mode — experimenting with completions, reading docs, and forming first impressions of what AI assistance can do. The real acceleration starts when the first habits form.
Your role: Observer
The bottleneck: Awareness — agents aren't yet part of the daily habit loop
Engineers are actively using Augment suggestions and building habits around AI assistance. Completions flow regularly; the team is learning to recognize where to lean on the agent and where to stay hands-on. The next step is trusting the agent with larger blocks of work.
Your role: Habit-builder
The bottleneck: Trust — engineers still review every suggestion before accepting
Phase 2 of 4
Delegating inside the IDE
Agent actions are live in the IDE. Engineers are starting to delegate larger tasks and reviewing at the block level rather than line by line — a real shift in rhythm and velocity is beginning to take hold. The IDE sandbox still defines the boundary.
Your role: Delegator
The bottleneck: Review overhead — delegation exists but line-by-line habits persist
A meaningful share of engineering work now flows through Augment. The team has developed its own AI-assisted patterns and is growing genuinely comfortable with agent-led development. Agent use is widening across more engineers and more task types.
Your role: Partner
The bottleneck: Reach — agent use is growing but not yet the majority workflow
More than half of engineering work involves Augment agents. The IDE-first workflow is well-established; the team is fluent in agent-led development and approaching the threshold where CLI-based work starts to unlock compounding gains. The IDE ceiling is becoming visible.
Your role: IDE-first engineer
The bottleneck: Channel — IDE-only limits the horizon; the CLI unlock is next
Phase 3 of 4
Escaping the IDE sandbox
Engineers are running Augment from the CLI for longer-horizon tasks, spinning up parallel agent sessions, and multiplexing between them — the first step beyond IDE-only workflows. This is where compounding productivity starts: tasks run concurrently rather than sequentially.
Your role: CLI multiplexer
The bottleneck: Coordination — parallel sessions multiply but switching costs rise
Multiple concurrent agents are the norm. Engineers are multiplexing between sessions, capturing compounding gains that IDE-only teams simply cannot match. The question is no longer whether to run agents in parallel — it is how many, and how to hand off context cleanly.
Your role: Parallel-agent operator
The bottleneck: Context — multiplexing grows but manual handoff between agents is costly
Ten or more concurrent agents are being managed at once, handling complex coordination by hand. This is the frontier of hand-managed orchestration — and the gravitational pull toward building a custom coordination layer is becoming real.
Your role: Multi-agent coordinator
The bottleneck: Scale — 10+ concurrent agents demand bespoke handoff protocols
Phase 4 of 4
Building AI infrastructure
Engineers build custom orchestrators that coordinate fleets of agents autonomously. This is AI infrastructure, not just AI tooling. Capacity scales without headcount limits — this team is defining what the next generation of engineering looks like.
Your role: Orchestrator
The bottleneck: Infrastructure — custom orchestration is the ceiling; fleet scale is next
No team is here yet. This is the theoretical frontier: org-wide agent swarms running autonomously, "dark factories" where code ships without human initiation, and agentic collaboration that extends beyond the software development lifecycle into product, design, and operations. The patterns don't exist yet — they're being invented by the teams at Stage 8 today.
The framework is not a report card. Stage 0 is not failure; Stage 8 is not the goal for every team. The goal is to understand where you are, what's blocking the next unlock, and whether that unlock is worth pursuing given your team's context and ambitions.
Further reading