Ready to become a certified watsonx AI Assistant Engineer? Register now and use code IBMTechYT20 for 20% off of your exam → https://ibm.biz/BdpBwE Learn more about AI Code-Generation Software here → https://ibm.biz/BdpBwX Is AI-assisted coding the future? 🤔 Cedric Clyburn explores spec-driven development, a game-changing approach that combines LLMs with software development best practices. Learn how it differs from vibe coding, integrates SDLC principles, and improves coding workflows with requirements-driven precision. 🚀 AI news moves fast. Sign up for a monthly newsletter for AI updates from IBM → https://ibm.biz/BdpBwH #aicoding #llms #aiassisted
ADVERTISEMENT
Anybody in the software development world ever heard of Software Requirements Specification (SRS)?
So we are going back to waterfall 👀
Writing and reviewing code was never the hardest part. It's always been knowing what to build and staying the course to do it right. Coding it is the easy part.
Back when humans were writing code, it seemed impossible to get specs/requirements. "Just implement something" I was told with some vague hand-waving, "if we don't like what you make you can just do it again". Now that machines are writing the code I'm being told "you need to give the poor LLM a proper spec so it can do its work". Damn machines get more respect than I ever did as a dev.
Writing and reviewing code never was the hardest part. People who think that failed before, and will continue to do so.
TDD with Gherkin specifications is back, classic!
Nobody talk about his writing backward?
Good video. These bitesized IBMs vids are great. But Spec Driven Dev has a few flaws: 1. it means someone has to write a spec, ie decide what to build up front. 2. LLMs are non deterministic, regardless of how good your prompt is. When you iterate on your v1 spec (ie. via user feedback) it’s going to rebuild your app and be potentially different for users. 3. there’s no feedback loop in the build, that’s useful to validate the design. It’s why everyone stopped with waterfall. Small specs work. But rarely create mega value. We’re having to modify our working practices to suit the current flaws of LLMs rather than doing what’s right for users. It’s also why there’s a drought of new good ideas, despite it being easier than ever to build. We have to find something better than SDD soon.
IBM has been steadily releasing these incredible learning videos and I'm rarely disappointed in their content. This is no exception, good work and thank you!
Spec-driven dev finally explains why my “vibe coding” sessions feel like arguing with a very confident intern. 😂 Loved how you tied specs, SDLC and LLMs into one clear mental model—this is exactly how I’ve been structuring my own AI-assisted workflows too.
I am amazed by how he could comfortably mirror-write and draw while maintaining a neat layout. 🤯
Actually, writing and reviewing the code was always the easiest part. Hardest part has always been coming up with a good design that meets the requirements, but that is flexible enough to match how requirements change over time. This isn't just infrastructure, but also the structure of the code itself - because in order to review it, it needs to be readable.
This is exactly what we (our lead guy) are doing at our organization.
The other advantage of this approach is often in regular software development they end up skimping on the documentations where here one can't do that in that it comes first and needs to be good enough to inform the AI how to wrote the code.
Great idea, done this way for military projects my entire career. Leaf level testing is critical as well.
We've started prod testing SDD tools at our org a week ago, it feels great to be driving these initiatives and your videos help strongly reinforce them !
As a software engineer I was playing with Claude code, then I noticed that it does better when we separate planning and execution, then it does even better when we first design, then plan, then implement and so on, turns out I independently invented spec-driven development :)
Utilising TLA+ in specs dramatically increased my velocity. It not only provides me the requirements, but also tests to eliminate entire classes of bugs.
I disagree with the traditional coding premise at 8:58. Specs came before code, not the other way around, at least in good software engineering organizations. Developing with AI has finally evolved past Prompt Engineering and Context Engineering to Specification Driven Development, which is another way of saying “good software engineering”.
I've been doing spec driven for new development for a while and have really good contract templates. Finally releasing complete, secure, compliant code. I have not figured out spec driven development for legacy app modernization. I'm happy to hear one reputable person touch on how the sdlc changes.