AI doesn't just change how software is built. It changes how projects are delivered.

Over the past eight months, we've used AI agents as part of a large eCommerce transformation. The technology changed the pace of delivery almost immediately. It also changed how teams worked together, how quickly decisions were made and where experienced people added the greatest value.

What surprised us wasn't how much changed. It was how much didn't.

The fundamentals of successful delivery - clear objectives, experienced teams, disciplined governance and engaged clients - became even more important as delivery accelerated.

Looking back, five changes had the greatest impact on the way the project was delivered.

1. Delivery becomes faster, but the bottleneck moves

The most obvious change is speed.

Once teams had adapted to new tools and ways of working, development capacity increased by around three times after an initial three-week learning period.

That didn't make the project easier to manage.

As delivery accelerated, the focus shifted from production to planning, validation and decision-making. The challenge was no longer writing software, it was ensuring the right software was being built at the right time.

Success became less about the volume of code produced and more about business value, quality and lead time.

The working day changed too. Rather than working continuously with AI, teams settled into focused delivery sessions followed by technical reviews and business validation. Reviewing AI-generated work requires concentration and judgement. The workload didn't become lighter, it shifted.

2. AI changes the work - not the people

One of the biggest misconceptions about AI is that it replaces technical roles.

In practice, every role remained essential. What changed was where people added value.

• Developers spent less time writing code and more time reviewing AI outputs and making technical decisions.
• Business Analysts used AI to accelerate documentation but remained responsible for ensuring requirements were complete and accurate.
• Technical Experts defined the boundaries within which AI operated and maintained technical quality.
• Project Managers coordinated AI-assisted delivery while keeping teams and clients aligned.

As development accelerated, a single Business Analyst could no longer support the team effectively. Additional Business Analysts were needed to define requirements, answer questions, and keep decisions flowing so developers could maintain momentum.

AI reduced the time spent on routine production tasks, shifting the focus of the role towards judgement, prioritisation, and decision-making.

3. Faster delivery demands faster decisions

Traditional agile practices remained important, but they were no longer enough on their own.

As delivery accelerated, we introduced:

• shorter synchronisation meetings;
• client validation every 48 hours;
• demonstrations once or twice a week; and
• closer collaboration between engineers and Business Analysts before implementation.

The team refined its approach as development progressed. Earlier collaboration clarified requirements before development began, allowing developers to focus on implementation with fewer interruptions and less rework.

The biggest change wasn't technical. it was organisational.

Faster delivery created more decisions, more opportunities for feedback and a greater need for teams to stay aligned.

4. AI needs context to be effective

Successful agentic delivery depends on more than access to AI tools.

AI performs best when it operates within a structured delivery environment, built around:

• a shared knowledge base that evolves throughout the project;
• specialist AI agents responsible for different delivery tasks;
• reusable AI skills and standardised commands; and
• Model Context Protocol (MCP) connectors linking AI directly with tools such as Jira, Confluence and Figma.

Creating this environment required an upfront investment in documentation, governance and tooling.

Once established, however, it became a reusable asset that continued to support application maintenance and future projects.

5. Productivity comes later than you think 

One expectation surrounding AI is that productivity improves immediately.

That wasn't our experience.

The first few weeks were spent learning new tools, refining prompts and establishing review processes. Productivity improved as the team gained experience using them.

Once those new ways of working became established, delivery happened quickly.

The benefits were not immediate. They emerged as the team became more familiar with the tools, refined prompts and embedded review processes into everyday work.

Five things that don't change

AI changes the pace of software delivery, but it doesn't change the fundamentals of successful projects.

If anything, it makes them more important. As development speeds up, unclear requirements, slow decisions and weak governance become apparent much sooner.

1. Clear project scope still matters

AI performs best when objectives are clear.

Business priorities, technical constraints and delivery milestones still need to be agreed before development begins. AI can generate convincing outputs from incomplete information, but it cannot resolve uncertainty about what a project is trying to achieve.

Well-defined objectives, good documentation and agreed priorities remain the foundation of successful delivery.

2. Experience becomes more valuable

AI changes how people work. It doesn't reduce the need for expertise.

Throughout the project, experienced engineers, architects and Business Analysts were better able to challenge AI-generated outputs, identify subtle issues and judge whether a solution would work in practice.

For less experienced team members, the challenge was different. Reviewing AI-generated work requires technical knowledge, confidence and critical thinking. Rather than narrowing the skills gap, AI can make it more visible.

That makes mentoring and knowledge sharing more important than ever.

3. Strong governance remains essential

Project governance changes less than many organisations expect.

The tools evolve, but the principles remain the same.

Successful delivery still depends on:

• clear accountability;
• realistic planning;
• effective risk management;
• transparent reporting;
• disciplined budget and scope control; and
• regular communication with stakeholders.

AI introduces new operational measures, such as token consumption and AI-assisted activity, but these complement existing project management practices rather than replace them.

4. Clients need to be more involved

One assumption is that AI reduces the need for client engagement.

Our experience suggests the opposite.

As delivery accelerates, business decisions need to happen sooner. Features are ready for review earlier, priorities need to be confirmed more quickly and feedback has to keep up with development.

Regular client validation proved to be one of the strongest contributors to project success.

Faster delivery only creates value when decision-making keeps pace.

5. Quality standards do not change

AI doesn't make quality assurance optional.

Testing, security, accessibility and performance remain fundamental to software delivery, regardless of how much AI contributes to development.

Throughout the project we maintained the same standards, including:

• 70–80% automated unit test coverage;
• accessibility reviews integrated into every delivery cycle; and
• end-to-end testing from the early stages of development.

As delivery accelerates, these disciplines become more important, not less.

Three lessons for delivery leaders

Three observations from this project are likely to be relevant to other organisations adopting AI.

Plan for adoption

Teams need time to learn new tools and establish new ways of working. Building this learning period into project plans leads to more realistic expectations and stronger long-term results.

Invest in experienced people

AI reduces manual effort, but it increases the value of good judgement.

Business Analysts, Technical Experts and Project Managers become more influential as delivery accelerates. Strong requirements, sound technical decisions and effective coordination create the conditions in which AI can deliver real value.

Agree how the client will engage

Regular client reviews are part of the delivery model.

Before the project begins, agree how decisions will be made, who will provide feedback and how quickly priorities can be confirmed.

In summary

Much of the conversation around AI focuses on what the technology can do.

A more useful question is what happens once it becomes part of a live delivery team.

Our experience shows that AI changes the mechanics of software delivery. It changes the pace of work, how teams collaborate and where expertise is applied.

What it doesn't change are the foundations of successful delivery.

Clear objectives. Experienced people. Disciplined governance. Engaged clients. A relentless focus on quality.

Those principles mattered before AI became part of software delivery. They matter even more now.

FAQ

What is an agentic delivery project?

An agentic delivery project uses specialised AI agents to support activities such as code generation, documentation, specification writing and testing. These agents work under human supervision, allowing delivery teams to focus on decision-making, quality and business outcomes.

Does AI replace developers or project managers?

No. Roles evolve rather than disappear. Developers spend more time guiding AI and reviewing outputs, while Business Analysts, Technical Experts and Project Managers focus increasingly on context, quality, governance and coordination.

What benefits can organisations expect?

Once teams have adapted to new ways of working, AI can increase delivery capacity, automate repetitive tasks and shorten delivery times. The greatest benefits come when AI is combined with experienced people, clear documentation and effective governance.

How long does it take to see productivity gains?

Based on our experience, organisations should allow around three weeks for teams to adapt to new tools and working practices. Productivity typically accelerates once new workflows and review processes become established.

What are the prerequisites for success?

Successful adoption still depends on the fundamentals:

• clear project objectives;
• up-to-date documentation;
• experienced delivery teams;
• effective governance; and
• active client engagement.

AI strengthens these capabilities. It does not replace them.

Does AI reduce the need for testing and quality assurance?

No. Faster delivery makes quality assurance even more important. Testing, accessibility, security, performance and code reviews remain essential to delivering reliable software.