What will occur if the machines used to write the code decide to make their own choices? This is not something that can be ignored now that AI agents are no longer just assistants for code writing but are beginning to take part in the more complicated processes of development.
Developers are no longer dealing with AI as an additional tool by their side. They are trying to determine how much autonomy should be provided to these systems, how the work of AI can be checked and verified, and what human interference is still required.
The shift could make software development faster and more efficient, but it also creates uncomfortable questions about ownership, oversight, and responsibility. As AI becomes more capable, the industry must rethink what development actually means.
McKinsey’s Charlotte Relyea and Martin Harrysson think that gen AI’s killer app is software development.
They reason that AI agents already run complicated tasks. Our role is to test, evaluate, and react to agentic decisions. And it’s happening right now. They give an example of AI teams spending the night shift refining a new cross-border payment system.
The day shift begins with a human engineer pulling AI-generated data, testing evidence, and flagging risks. The software engineer’s job isn’t to code, but to use their judgment and structure agent tasks.
Both Relyea and Harrysson claim the approach’s payoffs include smaller teams, lower unit costs, and quicker idea-to-influence cycle times.
AI may make writing software quicker and cheaper, but without the right tools, the move from production to distribution and sales can create a bottleneck.
And on the SaaS front, selling software isn’t as straightforward. The company model decides how to manage subscription management. Selling software globally comes with another set of headaches.
PayPro Global suggests partnering with a third-party global payment platform or payment gateway specializing in global payment solutions. Taxes, compliance, software payment processing, and fraud threats are issues many SaaS companies struggle with.
Invest in a decent payment platform, and these problems fall by the wayside.
“We overestimate technology in the short term and underestimate technology in the long term.” — Arthur C. Clarke.
Bain & Company argues that many corporations have AI tools, yet few have the engineering systems supporting them.
That’s where the discourse comes in. Others in the field agree: to lead in AI, businesses need the right technology and systems to turn AI’s potential into real outcomes.
Previously, the biggest challenge for AI-assisted software development was the limits of AI models. That’s changed. Today’s leading AI models can reason, write code, and handle complex development tasks.
However, the greater challenge now is the technology connecting AI to the rest of the development process.
Bain’s latest Tech and Engineering Survey shows that businesses expect software release cycles to become 148% faster and developer productivity to improve by 95% over the next one to two years.
That’s a big jump from the 20% to 27% productivity gains most companies are seeing today.
Embedded Computer Design’s white paper on the verification void created by AI-generated code brings in the trust factor.
It examines the challenge of checking AI-generated code. It also demonstrates why AI reviews, testing, and traditional code analysis cannot show the same level of certainty as formal verification.
One solution is AI-assisted formal verification, which can make thorough software testing more practical and scalable. Teams can then find more bugs, reduce false alarms, and be confident about the code they release.
| Metric | Figure | What it Shows |
| Senior technology leaders surveyed by Bain | 293 | Size of Bain’s latest Tech and Engineering Survey |
| Expected improvement in software release-cycle speed | 148% | Increase companies expect over the next 1–2 years |
| Expected improvement in developer productivity | 95% | Productivity increase companies anticipate over the next 1–2 years |
| Current productivity gains | 20–27% | Gains companies are currently achieving across key metrics |
AI isn’t replacing software development so much as changing what the job looks like. As AI agents get better at writing and executing code, developers are moving towards roles that demand more judgment, oversight, and verification. Building the software is only one part of the equation.
Companies also need the infrastructure to test, secure, distribute, and sell what AI helps them create. From formal verification to global payments and subscription management, the systems surrounding AI will become as essential as the models themselves.
The real question isn’t whether AI can build software, but whether companies can build the right systems around it to make that software trustworthy.
Ans: AI is taking on more complex software development tasks, including reasoning, writing code, and running agentic workflows. The developer’s role is shifting to testing AI outputs, evaluating risks, and managing AI agents’ work.
Ans: The challenge is less about AI model capabilities and more about the systems AI integrates into the wider software development process. Companies need the right engineering infrastructure to turn AI capabilities into reliable results.
Ans: AI-generated code can introduce bugs and other risks. AI reviews, testing, and traditional static analysis can help identify problems. But formal verification can provide stronger mathematical guarantees.
Ans: Global software sales can involve subscription management, taxes, compliance, payment issues, and fraud risks. Third-party payment and commerce platforms can help SaaS companies manage these challenges as they expand into international markets.
