A growing pain which I have noticed while working with a lot of SaaS vendors is their TAT on feature requests or bug fixes. You raise a ticket, get on call, explain the issue, sometimes explain again to their product teams, their engineering team gets involved, they negotiate on timelines and in depending on the scope and requirements it can take 1-2 months for the delivery, and even longer for enterprise software with extremely long release cycles, like once in 6 months.
It’s understandable, they have multiple customers with various requirements, their own internal product roadmap and of course engineering tech debt to clear.
But deep down there are more reasons, sometimes they may lack the technical expertise to implement the feature, sometimes they have expertise but not enough resources. Sometimes their org structure has completely changed, the products you rely on are deprioritised.
While I can empathise with all this to some extent, but these days my patience is running thin.
“If only I can connect my Claude to their codebase”
That’s the feeling I get, and I understand it ignores the complexities involved of a typical SDLC, but that’s the feeling I get.
Instead of spending multiple hours explaining my use-case on zoom calls, listing my requirements, negotiating priorities, taking screenshots, recording videos, reviewing their proposals, creating the reproduction steps for their support agents; why can’t I just contribute to their code directly ?
I believe if your developing SaaS in 2026 and you want it to remain relevant in 2036, you should do this: Expose an Agent (via a CLI or MCP whatever) which I can connect from my AI harness.
My harness already has all the information (in most cases) around the feature needed or bug encountered, because in most cases that discovery has been made through the harness (if you use the harness correctly).
The harness can then communicate with the SaaS-“Agent” and instruct it to make the required changes. In an ideal world, the SaaS-Agent goes through the entire SDLC of building that feature (could just be a Claude on the other side). It could very well be scoped to just me, why do we need to go to entire QA cycles for a bug or feature raised by me ? My harness can test it out, if it doesn’t work, it’s okay, my harness will instruct your SaaS-Agent accordingly.
It can and it should in the future, also consider the other conversations it is having with the other customers, if they mention the same thing, just enable the feature for them and ask their agent to confirm it works for them. Isn’t this true agent-to-agent communication that people want ?
I understand this is a moonshot, but even opening a PR is a lot. So much time goes into negotiating timelines with the SaaS vendors, once there is a PR, it’s just a matter of reviewing them (using AI of course 😛 )