Every few weeks there’s a new headline declaring that AI is going to replace developers, kill agile, and make project managers obsolete. I’ve been in delivery long enough to recognize a panic cycle when I see one.

Here’s my actual take after watching teams try to adopt AI in practice: it’s not replacing the SDLC. What it is doing - when teams use it well - is attacking the part of software delivery that quietly drains everyone’s energy: the overhead.

And there is a lot of overhead. More than most leaders want to admit.

Regardless of whether your org runs waterfall, agile, or a hybrid - the inefficiencies I see aren’t in the coding itself. They’re in the gaps between the work:

• Requirements that get misunderstood because no one had time to fully clarify them
• Decisions made in a meeting that never made it into a ticket
• Tickets so vague the dev has to ping three people before they can start
• QA catching things that should have been caught in requirements
• Teams stuck waiting on handoffs while the sprint clock ticks
• Documentation that was accurate six months ago and now actively misleads people
• Leaders asking for status updates that take too long to compile

That’s where AI can make a real difference. Not by writing your code for you, but by shrinking the operational drag that surrounds the actual work.

AI in Waterfall: Less Document Wrangling, More Thinking

Waterfall gets a bad reputation, but I’ve worked in regulated industries and large enterprise programs where it’s simply the reality. The governance requirements are real. The documentation requirements are real. And the amount of time teams spend manually transforming deliverables from one phase into the next can be painful.

That’s where AI starts to shine in waterfall environments. Things like:

• Converting business requirements into technical specs (instead of a BA doing it alone at 9pm)
• Generating traceability matrices that could otherwise take days
• Drafting test scenarios directly from requirements
• Summarizing 80-page requirement documents, or workshop meeting notes for stakeholders who won’t read 80 pages
• Flagging conflicting requirements before they become expensive change requests

I want to be clear about something here though: AI doesn’t replace the business context. On large programs with distributed teams, the nuance of what a business actually needs still has to come from people who understand it. AI can accelerate the translation of that context into documents. It can’t supply the context itself. That’s still a human job - and a very important one.

AI in Agile: Giving Teams Back Their Time

Agile was supposed to reduce bureaucracy. And it does, in theory. In practice, agile teams can spend more time managing their process than is needed or wanted.

Grooming sessions that run long. Acceptance criteria rewritten three times because the first draft was too vague. Sprint retros that produce no actionable output. Jira that’s perpetually out of date. Release notes that someone has to write manually at 5pm on a Friday.

None of this is unique to any one team. It’s structural. And AI can take a meaningful bite out of it:

• Generating user stories from discovery notes so the team starts grooming with something real
• Drafting acceptance criteria that a product owner can edit rather than write from scratch
• Summarizing retrospectives and surfacing patterns across sprints
• Identifying delivery risks from ticket aging and sprint velocity trends
• Producing stakeholder updates in minutes instead of hours

The goal isn’t to automate agile. It’s to give people back the time they’re currently burning on administrative work so they can do the things that actually require judgment.

Hybrid Organizations Might Have the Most to Gain

Here’s an opinion I’ll stand behind: most organizations are not actually doing “pure agile.” They have executive roadmaps. They have quarterly commitments. They have governance boards and milestone reviews and cross-functional dependencies that agile ceremonies alone can’t account for.

That hybrid reality - agile teams operating under waterfall-ish planning cycles - creates its own overhead. Translating sprint progress into executive language. Connecting implementation work back to roadmap objectives. Tracking dependencies across teams that operate on different rhythms.

AI can help bridge that gap in ways that used to require a small army of PMO staff: summarizing delivery health across portfolios, correlating defects back to requirement gaps, improving forecasting using historical sprint data. That’s where AI stops being just a code assistant and starts becoming a genuine delivery accelerator.

The Bottleneck Usually Isn’t the Code

Code generation gets all the press. But in my experience managing large programs with dozens of stakeholders - product, engineering, QA, offshore teams, architects, infrastructure, business owners -  the code is rarely the actual bottleneck.

The bottleneck is usually communication. Alignment. Someone waiting on a decision that got buried in a Teams thread. A ticket sitting in review because the acceptance criteria were ambiguous and nobody wanted to guess wrong.

When a meeting ends and AI automatically produces a summary, extracts action items, links them to existing tickets, and flags blockers - that’s not a small thing. That’s hours of coordination work that no longer falls on a single person to do manually.

AI Still Needs Someone Steering

I’d be doing a disservice if I made this all sound frictionless and easy. It’s not.

The organizations I’ve seen struggle with AI adoption usually have one of two problems: they trust the outputs too much, or they use AI to move faster in the wrong direction. Bad requirements generated faster are still bad requirements. A poorly scoped ticket auto-generated from vague notes is still a poorly scoped ticket.

The risks are real:

• Technical debt from AI-suggested solutions that weren’t properly reviewed
• Security gaps from generated code nobody fully audited
• Compliance issues when AI output is treated as final without expert review
• False confidence from polished-looking documents built on wrong assumptions

The teams that get the most out of AI aren’t removing humans from the process. They’re pairing experienced judgment with AI efficiency - letting AI handle the volume, letting people handle the thinking.

Final Thought

The SDLC isn’t going anywhere. Agile isn’t going anywhere. Developers and delivery managers aren’t going anywhere.

What I hope does go away is the manual overhead that surrounds software delivery without actually making it better. The status reports nobody reads. The documentation that’s outdated before it’s published. The hours spent reformatting information from one format into another.

That’s the real opportunity — not replacing expertise, but finally giving people enough breathing room to use it. And if it helps chip away at the technical debt that’s been quietly accumulating since 2019? I’ll take that too.

At AAXIS, we’re seeing this firsthand across enterprise delivery organizations. The companies getting the most value from AI aren’t trying to replace their SDLC — they’re using AI to reduce operational friction, improve visibility across teams, and accelerate execution. From intelligent documentation workflows to operational data harmonization and AI-enabled delivery support, the real opportunity is creating environments where teams spend less time managing process and more time delivering impact.