I’m curious to know if I’m the only one who feels like the odd one out for saying that we can’t build truly complex agentic systems using purely no-code or low-code approaches.
Lately, platforms like AgentForce and AgentSpace seem to be fueling the idea that you can build and deploy any kind of agency system with little to no coding. But I don’t think that’s realistic. Building agentic systems isn’t just about stringing together prompts or deploying with a few clicks—it requires thoughtful design around fault tolerance, resilience, and system architecture.
For instance, what happens if your agency system fails at the third agent? Do you start over or resume from a checkpoint? What about SSO integration, or steering agent behavior based on user roles? And how do you handle ecosystem-level context propagation to agents?
These are not things you can reliably do with prompt-based coding alone.
You’re not alone, but none of this stuff is really new either.
It seems like it’s a typical “in a gold rush, sell tents, pants, and pickaxes” type of situation.
I think the best you (or we all) can do is remind people how compound probability scales before they heavily invest in these things. Manage the expectations of people you care about.
That said, I don’t think it’s bad if people play with these tools so they get an understanding of the challenges involved and opportunities possible in building these things.
I am not telling how - but I say it is possible to automate that. And if you think about it and try some stuff for a year or two too you will also find a solution for it - that for sure.