From Hype to High-Value: How We Are Re-Engineering Software Delivery with Agentic AI. My perspective as a Crosstide Tech Lead">
I’m still working through what good looks like with AI in software delivery, but a few things are becoming clearer.
My usual car analogy still works best. AI is like adding more horsepower. Useful, but not if the brakes, steering and driver are not up to it.
In software terms, that means judgement, patterns, standards, testing and delivery discipline still matter. Probably more than before.
The teams getting value from AI are not just asking it to write code faster. They are using it to sharpen the way they already work. Challenging requirements, exploring options, spotting gaps, generating test scenarios, improving documentation and taking some of the pain out of repetitive delivery tasks.
Where I have seen it work well is when the team already has a clear engineering approach. They understand the domain, have sensible patterns, review properly and test properly. In that environment, AI can be a genuine accelerator because it has something solid to work within.
Where it becomes risky is when AI is used to cover up weak delivery practice. If the requirements are vague, the architecture is unclear or the test coverage is poor, AI can simply help produce more output faster. That does not automatically mean better outcomes.
The main things I’m still learning are:
Senior engineering judgement matters more, not less
AI can help produce code, options, test scenarios, documentation and ideas very quickly. But someone still has to shape the approach, set the patterns, challenge the assumptions and decide whether the output is actually good enough.
That judgement matters. Is the approach maintainable? Does it fit the architecture? Is it secure? Will another engineer understand it later? Does it solve the actual business problem?
This also changes the shape of the team. AI doesn’t remove the need for engineers, but it does increase the value of smaller, more experienced teams who can make good decisions quickly. People who understand the domain, can challenge the output, and can tell the difference between something that looks right and something that is right.
A faster answer is not automatically a better answer. AI can accelerate thinking, but it cannot take ownership of the decision.
AI is useful across the SDLC, not just in code generation
Some of the more useful areas are before and around the coding itself. Turning rough thinking into clearer requirements. Challenging acceptance criteria. Spotting missing scenarios. Pulling out assumptions. Making documentation less painful to keep current.
Used well, AI can act as a challenge tool. A second opinion. A way to pressure test thinking before work gets too far down the line.
But teams need to be careful. AI can make uncertainty sound more certain than it really is. A polished answer is not always a complete answer.
Testing, review and evidence become more important
If AI helps teams move faster, the controls need to keep up. Better technical review, better automated testing, clearer regression coverage, more focus on edge cases and better evidence of why decisions were made.
It also means being clear about where AI has influenced the work. Was it used to generate code, refactor an approach, draft test cases, review requirements or produce documentation? None of that is necessarily a problem, but it should not be invisible.
Clear standards. Good patterns. Proper testing. Sensible documentation. Human ownership.
None of this is new, but AI makes the gaps harder to ignore.
Teams need to get the basics right before they chase the bigger, mostly unrealised AI gains.
Once the basics are in place, AI can absolutely help teams move faster. But without them, it is just more horsepower with drum brakes.
That is the thinking behind our AI SDLC review at Crosstide. We look at where AI can genuinely help across the software delivery lifecycle.
If you are looking at AI in software delivery and want a practical view of where to start, get in touch or take a look here: https://