Blog / Other resources

What are the levels of AI adoption in software engineering?

The 5 levels of AI adoption in software engineering, from human-only to fully autonomous agents, and how to find your level from the work your team ships.

What are the 5 levels?

Level Name Who does the work
L1 Human-only Humans write all the code
L2 AI-assisted Engineers write with AI tools like Copilot or Cursor
L3 AI reviews Humans write, AI checks the work
L4 Orchestrated agents Agents carry out whole tasks, humans direct and review
L5 Fully autonomous agents Agents start and deliver work, humans govern

How do you find your level?

  1. Connect the tools the team already uses: code, tickets and agent sessions.
  2. Measure who delivered each piece of finished work: an engineer, an AI-assisted engineer or an agent.
  3. Read the share of delivered work at each level, month by month.

A survey tells you which tools people use. The delivered work tells you which level the team is really at.

Why does the level matter?

  • Budget: at Level 2, many teams pay for AI tools before they see the return.
  • Pricing: at Levels 4 and 5, hours stop being a good unit for a quote, because agents do whole tasks.
  • Quality: as the level goes up, review becomes the main control.

How Pensero helps

Frequently asked questions

What is the most common level?

Level 2, AI-assisted. Engineers write the code with AI tools.

What is the agentic AI maturity model?

It is the same 5-level scale. Levels 4 and 5 are agentic: agents carry out or start whole tasks.

How do I measure my team level?

Measure the delivered work, not the tool licenses. Pensero scores each piece of finished work by magnitude × complexity and shows who delivered it.

What do I do after I know my level?

Decide what to improve next, for example review or agent use. Then measure the delivered work, the cost per delivery point and the quality again.