PhoenixDX Blog | Software Development & Low-Code

What is your team's AI Engineering maturity? And why does it matter?

Written by Patricia | Nov 19, 2026, 6:13:01 AM

Wondering why your team isn't getting the AI results your business needs? The issue may not be the AI models or tools they're using, but the level of AI engineering maturity your organisation has to apply them effectively.

This article explains what shapes AI engineering maturity, what each level looks like, what it means for your team's capability to build and run software — and why Level 4 is the level worth aiming for.

The determinants of AI engineering maturity

Four key measures shape how mature a team is at using AI to build software. Each looks at a different part of the picture, and together they decide how much work AI can reliably take on — and how much you can trust what comes back.

Foundation is what AI knows, and how the team is set up around it. AI can only work with the information it has, so this looks at how well it understands your systems and business rules, and how people and AI share the work.

Governance is the rules AI works within. It covers what AI can access, how its work is checked for security issues, and the design standards it follows, so you can use AI more widely without adding risk.

Delivery is how AI takes part in the work. It ranges from occasional help with a single task to AI working directly on your systems, and it shapes where the bottlenecks sit.

Outcomes/learning is how you know it's working. It looks at how AI's work is tested, whether lessons are captured so mistakes aren't repeated, and whether the process keeps improving over time.

None of these measures stands alone. Fast delivery without governance creates risk, and strong governance without good foundations limits what AI can do. As a team develops all four together, its use of AI matures in recognisable stages.

The levels of AI engineering maturity

Those stages are the four levels of AI engineering maturity. Each level reflects how far the four measures have developed — how much context AI has, how clear its rules are, how deeply it takes part in the work, and how well its results are checked — and what that makes possible for your team.

Each is a legitimate stage with its own benefits; each lays the groundwork for the next, and Level 4 is where AI starts to deliver its full value. Most teams sit at different levels across the four measures, so treat these as a guide rather than a single score.

Level 1 — No AI process

AI isn't part of how software gets built. Some staff may still use AI on their own, without guidance.

What this means for capability: Your team is not yet benefiting from AI, even informal AI use.

Level 2 — Chatbot-assisted

Developers use an AI chat tool, copying snippets in and answers out. The team and process stay the same: AI helps with specific tasks such as drafting tickets or queries, a few basic rules sit in prompt files, and a person scans the output for obvious problems. This is where most businesses are today.

What this means for capability: Because the wider process hasn't changed yet, team-level delivery tends to move only a little, and results vary with each person's experience of the tools. It's a valuable stage for building familiarity and confidence.

Level 3 — Human-in-the-loop

AI agents work directly in your codebase, and a developer reviews every line. Someone maintains a context file that tells the AI about your systems, AI's work runs through your existing security scanning, and people enforce design boundaries. Teams become leaner, use agents daily and hand them work to do in parallel.

What this means for capability: This is where team capability starts to shift. Developers can run several pieces of work in parallel, try alternative approaches cheaply, and tackle improvements that never made it up the backlog. The natural limit is review time: when every line is checked by hand, people can only verify so much.

Level 4 — Systematic harness

The process itself is built for AI. Your code repository becomes the single source of truth the AI draws on, design standards are written as automated checks, AI access is scoped and auditable, and AI reviews other AI work against quality criteria people set. Small generalist teams own each initiative, agents run continuously, and every failure feeds back into system improvement.

What this means for capability: Delivery is no longer capped by how much people can review by hand. Small teams can look after more systems, quality stays consistent because checks are built in, and every issue that slips through becomes a new automated check, so capability keeps improving. This is where AI becomes a dependable part of how your business delivers software, rather than a tool a few people use well. Getting there takes deliberate investment, and the right support can shorten the journey considerably.

Beyond Level 4, there's an emerging fifth level where AI helps improve the development process itself, but it remains largely theoretical for now. That makes Level 4 the practical goal for any organisation that wants AI to make a lasting difference to how it builds and runs software.

So, what's your team's maturity?

You don't need to be technical to get a good sense of it. Ask your team, or your technology partner, these six questions:

  1. How does the AI learn about our systems and business rules?
  2. What can the AI access, and is that access limited and logged?
  3. How is AI-written work checked before it reaches customers — by people, by automated tests, or both?
  4. When the AI gets something wrong, what changes so it doesn't happen again?
  5. Who is accountable when AI-written work causes a problem?
  6. Where are we strongest, and where is our biggest opportunity?

Clear, specific answers usually point to a more mature approach. Less certain answers simply show where there's room to grow — which is exactly what you want to know before deciding where to invest.

Where to start

Level 4 is the destination, and every step towards it counts. Where you begin depends on where you are today:

  • At Level 1: agree which AI tools are approved, and what company data can and can't go into them.
  • At Level 2: bring AI into real work on a low-risk system, with your existing testing and security checks applied to everything it produces.
  • At Level 3: notice what reviewers keep catching by hand, and turn those into automated checks.


How PhoenixDX can help you reach Level 4

Building Level 4 practices from scratch takes time, specialist skills and a willingness to change how work gets done — all while delivering the projects your business already depends on. We know, because we've done it ourselves.

Our Vietnam AI-First Hub: Level 4 in practice

When we established our AI-First Hub in Vietnam, we didn't add AI to an existing way of working. We built the hub around Level 4 from day one.

That started with how we selected the team. We looked for engineers ready to work this way — people who could give AI the context it needs, judge what comes back, and turn every lesson into a better system. Then we structured the team around AI, building the harnesses they work within: the context their agents draw on, the guardrails that scope and log what AI can access, the automated checks that verify its work, and the feedback loops that turn every issue into a lasting improvement.

The result is a team that works at Level 4 every day. And it's the same team, with the same harnesses, that we now make available to organisations across Australia.

Two ways to work with us

Native-AI pods. A pod is a small, experienced delivery team from our AI-First Hub, built around Level 4 ways of working. Context, guardrails, automated checks and feedback loops are part of how the pod operates from day one, so your organisation can start benefiting from Level 4 delivery without first having to build it all yourself. Pods work alongside your people rather than around them. As they deliver, they bring your team into the same practices, so the standards, skills and know-how stay with your organisation after the engagement ends.

Consulting. If you'd rather build the capability in-house, our consultants help you assess where your team sits across the four measures, then put the foundations, governance, automated checks and feedback loops in place to lift your maturity — drawing directly on what we learned building our own hub.

Wherever you're starting from, we can meet you there: helping a Level 1 or 2 team put the right foundations and governance in place, or helping a Level 3 team turn manual review into a scalable system.

The businesses getting the most from AI aren't the ones with the flashiest tools. They're the ones that have built the right system around them. Knowing your maturity is where that starts — and if you'd like to explore what reaching Level 4 could look like for your organisation, the PhoenixDX team would be glad to walk you through it.

Let's Talk

This article draws on The Harness Model maturity model by Maciej Laskowski and Tomasz Michalak.