I’ve noticed that when we integrate generative AI into our software delivery cycles, stakeholders keep asking for "just one more feature" because they think AI makes coding instant. How are you all handling this specific type of scope creep without killing the team's momentum or the project budget?
3 answers
In my experience at a FinTech firm, the best way to handle this is by implementing a strict Change Control Board (CCB) specifically for AI features. We found that stakeholders often underestimate the "hidden" technical debt and compute costs associated with "small" AI tweaks. I always present a trade-off matrix: if we add this prompt-engineering task, we must deprioritize a core UI element. By making the cost of "instant" AI visible, you regain control over the project scope. Transparency is your best friend when dealing with hyped technologies like LLMs
Have you tried using a "Value vs. Complexity" scoring model for every new AI request that pops up mid-sprint? It usually helps stakeholders see that not every AI feature is worth the disruption
We moved to a "Fixed-Scope, Variable-AI" model. We lock the core features and treat AI enhancements as a separate experimental backlog that only gets touched if we have a surplus in the sprint
I totally agree with Linda! Separating the 'must-haves' from the 'AI-fluff' ensures the project remains viable even if the AI implementation hits a technical snag or takes longer than expected.
Robert, that scoring model is a lifesaver. I actually started using a 1-10 scale where anything above a 7 in complexity requires an immediate budget extension approval. It shifted the conversation from "can we do this" to "is this worth the extra $10k in API tokens and dev time." It really put the brakes on the "just one more thing" culture we were struggling with last quarter.