Blog / Other resources

What is the agentic SDLC, and how do you measure it?

The agentic SDLC is software delivery where AI agents do part of the work. Learn the 5 levels and how to measure what agents and engineers deliver.

How the agentic SDLC is different

Classic SDLC Agentic SDLC
Who writes the code Engineers Engineers, assistants and agents
Who reviews it Engineers Engineers and AI reviewers
What costs money Salaries and tools Salaries, tools, tokens and agent runs
What you must know Speed and stability (DORA) Speed, stability, and what each human and agent delivers for its cost

The classic metrics still apply. DORA metrics show how fast and how safely you ship. They do not show who or what did the work, or what that work cost.

The 5 levels of AI in engineering

Most teams move through 5 levels. The levels are a diagnostic, not a score: different work in the same team can sit at different levels.

  1. Human-only: people write and review all the code.
  2. AI-assisted: engineers write with tools like Copilot or Cursor and check everything themselves. Most teams are here.
  3. AI-reviewed: AI joins the code review.
  4. Orchestrated agents: agents take whole pieces of work, and engineers direct them.
  5. Autonomous agents: where the industry is heading.

How to measure the agentic SDLC in 4 steps

  1. Measure delivered work, not activity. Commits, lines of code, tokens and prompts show activity. Score each finished piece of work by its size and difficulty instead.
  2. Use one scale for humans and agents. If an engineer and an agent deliver the same change, it gets the same score. Then you can compare them.
  3. Connect cost to output. Add salaries, tool seats, tokens and agent runs, and divide by the delivered work. The result is the cost per unit of delivered work, for each team and each level. See how to measure the cost and output of AI coding agents.
  4. Track quality next to output. Watch rework, reverted changes and incidents. More output with more rework is not a gain.

Common mistakes

  • Counting pull requests or lines of code. Agents make many small changes, so these numbers go up without more value.
  • Counting AI seats or tokens as adoption. A paid seat is not delivered work.
  • Measuring agents apart from engineers. Then you cannot see if an agent is cheaper or better for a type of work.

How Pensero measures the agentic SDLC

The Delivery Point is one unit of finished work, scored by size and difficulty, the same way for engineers, AI-assisted engineers and agents. Pensero reads it next to cost and quality, for each team and each level.

Sources

Frequently asked questions

What does SDLC mean?

SDLC means software development lifecycle: the steps from planning to code, review, test, release and maintenance.

Is the agentic SDLC the same as AI-assisted development?

No. In AI-assisted development (Level 2), engineers write the code with AI help. In the agentic SDLC (Levels 4 and 5), agents do whole pieces of work, and engineers direct and review them.

Are DORA metrics enough for the agentic SDLC?

No. DORA metrics show speed and stability. They do not show what humans and agents delivered, or the cost per delivered unit.

How do I know which level my team is at?

Look at who does the work. If engineers write most code with assistants, the team is at Level 2. If agents deliver whole pieces of work, the team is at Level 4.

Which tool measures the agentic SDLC?

Pensero measures the delivered work of engineers, AI-assisted engineers and AI agents on one scale, the Delivery Point, with cost and quality for each team and level.