“As a software engineer watching the explosion of generative AI, I've noticed a widening gap between the headlines and the harsh realities of implementation,” says Paul Zietsman, Technical Director at cloudandthings.io. While tools like ChatGPT and Claude have captured everyone's imagination, it's time for a pragmatic discussion about what GenAI can and cannot do for your business.
The promise vs the practice
Let's be honest: GenAI is revolutionary. It can generate code, write content and analyse data faster than any human. But it's not the magical solution vendors might have you believe. As Andrew Ng recently pointed out: "There are a lot more opportunities than there are people working on them." But that doesn't mean every opportunity is right for your organisation.
What actually works today
From our experience in the trenches, here's where GenAI delivers real value
- Developer productivity: GenAI can handle about 20%-30% of coding tasks through code generation and documentation. However, it won't replace your development team – it augments them.
- Content operations: For technical documentation, internal communications and first drafts of materials, GenAI excels. But human review remains crucial.
- Data analysis: GenAI can help interpret data and suggest insights, but it needs clean, structured data and human oversight to be effective.
- Domain-specific knowledge processing: When properly focused, GenAI excels at making sense of large volumes of unstructured data within specific domains. For example, it can analyse thousands of technical documents, customer feedback or industry reports to extract patterns and insights that would take humans weeks to process.
The hidden costs nobody talks about
Let's start with the usual suspects: you'll need to budget for AI model APIs, compute resources, specialised engineering talent, training programmes and consulting services. Cloud costs can also spiral quickly, especially when dealing with large language models and high-volume data processing. But beyond these obvious expenses, there are several often-overlooked factors to consider:
- Data quality: GenAI needs high-quality, well-organised data to function effectively. Most organisations underestimate the cleanup required.
- Integration complexity: Connecting GenAI to existing systems isn't plug-and-play. It requires significant engineering effort.
- Ongoing management: Models need continuous monitoring and adjustment. This isn't a "set it and forget it" technology.
- Testing challenges: Testing GenAI applications is far more complex than traditional software. You'll need to account for variations in outputs, edge cases and potential biases – all of which can be time-consuming and resource-intensive to validate properly.
Practical next steps
If you're considering GenAI implementation:
- Start small with a pilot programme in a non-critical area.
- Focus on measurable ROI rather than technological sophistication.
- Invest in training your team – both technical and non-technical staff.
- Establish clear governance frameworks for AI use.
- Experiment with off-the-shelf GenAI tools first to understand how LLMs "think" and behave.
- Keep your pilot focused on a specific domain – avoid the temptation to build a "one bot to rule them all" solution.
The bottom line
GenAI is transformative, but it's a tool, not a magic wand. Success comes from understanding its limitations and focusing on practical applications that solve real business problems. As a technical leader, your role is to cut through the hype and focus on sustainable, value-driving implementations.
Remember, the goal isn't to have the most advanced AI – it's to solve business problems efficiently and responsibly, says Zietsman.
Share