Claude, Gemini, and GPT-5 can handle every SDLC task. Almost none of them should.
The New Stack Jeff Michael
Big AI models like Claude, Gemini, and GPT-5 can technically do every coding task, but the article says they shouldn't. The real risk now isn't bad AI output, it's companies building sloppy AI systems around it.
Based on reporting by The New Stack, Jeff Michael — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
The pitch for frontier AI models has always been their versatility. Claude, Gemini, and GPT-5 can plan, code, test, and review, all in one system. But the piece from The New Stack makes a sharp argument: just because a model can do everything in the software development lifecycle doesn't mean it should. Most of that work doesn't actually need frontier-level reasoning. It needs consistency and speed, which smaller, specialized models can deliver at a fraction of the cost.
The author leans on a Star Wars comparison to make the point land. Frontier models are the C-3POs of AI: broadly capable, but expensive to run at scale. Most SDLC tasks need an R2-D2 instead, something narrow and optimized for one job, whether that's generating unit tests, running build validation, or checking compliance. The framing is that software delivery doesn't need one brilliant assistant. It needs a team of specialists, each tuned to a specific stage of the pipeline.
That reframes the whole conversation. AI in production isn't a single model doing everything, it's closer to a supply chain: a planning model defining requirements, a coding model writing logic, a testing model building test suites, validation and compliance agents checking the work, and governance systems watching costs and performance across all of it. Frontier models still have a role, mainly bookending the process, helping define requirements up front or acting as an
My take — AI-written commentary, not fact-checked reporting
The comparison to early cloud adoption is the part worth sitting with. Companies rushed to the cloud, got hit with runaway bills, and then spent years clawing back control through hybrid strategies. AI is following the same script, except the stakes are higher because now the systems are writing the code too. Betting an entire engineering org on one frontier model because it feels impressive in a demo is exactly the kind of decision that looks fine in the pilot and painful on the invoice a year later.
Read more about this at: The New Stack