What the customer wanted, AI redux!
What the customer wanted – the old way
Some decades ago a cartoon went round the IT community showing “what the customer asked for”, the object in question being a swing, then what the business specified followed by what the developers built – and variations thereof, see Figure 1. The point of the cartoon was to show the miscommunication that occurs across the handovers and is typical in the world of IT. It was amusing and accurate about the world of IT. You can also think of the IT waves over the decades in terms of this cartoon: waterfall was a linear path from client to deliverable, big bang and often wrong due to little feedback; agile improved this with increments of minimal deliverables and plenty of feedback from client to developers in each iteration; DevOps automated delivery; platform engineering increased (and is increasing, being the current wave) infrastructure as code and developer autonomy. There is now a new wave to consider: AI-assisted software development.
Figure 1: What the customer wanted – traditional view:
What the customer wanted, with AI-assistance
AI can reduce the miscommunication gaps between stakeholders because it empowers business-oriented stakeholders to build applications, where the AI tool enables a new class of stakeholder to diretly build proof of concepts to full featured applications, such as web sites, eradicating miscommunication gaps. Let me expand on this. First, this is not the end of developers. There will be many types of applications where the expertise of developers will continue to be in demand, whether it is to adjust or expand a generated application, to applications that non-developers would not be able to even specify. Second, empowerment is crucial to business success, a lot of IT is wasted due to poor design, out of date delivery, and problematical code due to poor initial work that constrains later rework. AI-assisted software development short circuits old, cumbersome ways of working. For example, it delivers the agile goal of immediate realization of requirements, in seconds. The iterative process of refinement is accelerated, applications are built faster and with a better fit to the requirements. AI-assisted software development can improve the productivity of all stakeholders with a need to build applications.
There are caveats. As the recent DORA State of DevOps 2024 report showed, technical debt and code complexity can increase with the adoption of AI in development. This indicates the need to address the cultural aspects of AI adoption within an organization. For example, if executives stop hiring expensive experienced developers in favor of cheaper developer novices, expecting the AI to prop up the novices, then what happens is that the experienced developers stop writing code and spend more time reviewing the AI generated code – not the best use of their time. Productivity declines in this scenario.
Omdia recommends using dedicated AI-assisted coding tools rather than directly using a large language model (LLM) for development. At the time of writing, an LLM like ChatGPT should not be used directly for coding as it could easily be passing on many types of vulnerabilities, from coding errors to security holes, picked up during training. Purpose built AI-assistants are built with guardrails and minimize hallucinations and are therefore more reliable.
领英推荐
To see more of what the market offers in terms of AI-assisted software development see our upcoming Omdia Universe. To finish with a version of the original "what the customer wanted" cartoon but where I requested ChatGPT to inject AI assistance, see Figure 2.
Figure 2: What the client wanted – ChatGPT/Dall.E view
Further Reading