Agentic software development is trending, and it's easy to see why: you can generate working software in a matter of minutes just by typing a few phrases! Speculation is running wild. Are SaaS giants about to become history when anyone can allegedly vibe-code their own Slack or Jira?
No silver bullets in the enterprise
At the same time, senior developers who have lived through a few previous hype cycles already know that there are no silver bullets. Did someone actually vibe-code a better Jira replacement, and not just a prototype? Are they running it in production already? Is it actually usable, accessible, GDPR-compliant, and secure?
The enterprise Java systems we work with have been built over years or even decades. No two are alike, but they all have their own quirks, and many are big enough that no single person knows the entire codebase. Users have become accustomed to certain ways of working, and they're quick to let us know when something doesn't work the way they like. Business owners can have strict requirements about behavior, security, and compliance. Trusting your system's maintenance and development to an AI without constraints seems as untenable as rewriting everything in server-side JavaScript.
The fundamentals still matter
While traditional software engineering skills and practices may seem to be on their way to obsolescence, in reality they're needed just as much as ever. Creating a demo or a proof of concept with a single prompt is one thing; bringing agentic development into an enterprise environment requires the same old considerations: risk management, security awareness, and quality assurance. Software architecture still matters for the same reasons as ever: it's not just the quantity of features, but also all the necessary qualities of the system.
How to stay in charge
So how do you get the benefits of the agents' faster turnaround while maintaining control of your system's functionality and quality? You adopt a development process that keeps you in charge:
- Analyze your system to uncover unstated assumptions.
- Use the right tooling, such as relevant MCP servers and command-line apps, to reduce the risk of hallucinations.
- Verify everything with tests, so the agent doesn't get to rely on assumptions.
- Keep humans in the loop where needed, but let agents work independently as long as possible so they can focus on meaningful work.
- Do it securely, so you're not leaking confidential data or environment secrets.
Do it right, and you can increase the quality of your codebase, instead of turning it into spaghetti.
Get started
Sounds good? You can get started with Vaadin's intensive, one-week, hands-on training workshop, where you're coached by our experts and get to work on real features in your own system instead of simple demos. Our focus is on secure development of Vaadin-based applications, of any size and shape. With our help, you'll be delivering value in weeks rather than months.
Ready to try it on your own system? Read more and register for the workshop.