In my opinion, it is a little early to identify "success/failure" of LLM products just yet, especially Agentic. What we are seeing is the definition of hype, and maybe once that settles we will be able to see the reality a little better. From what I have seen, with the hype much less than early last years, is that customers are more cautious of just "AI labels" on product, but if it does solve a unique problem that was not possible to solve for earlier then its different. In your case, if you are saying that the company is trying to replicate a competitive product, just through Agentic, might not be a very good idea in the long run. Though it also depends on how they decide to build it further - i.e. start with replicating, but then a way bigger roadmap that your competitor couldn't catchup to just because they never built a base (because many are.
Author here. I've been wrestling with a core tension in building AI agents: the balance between rigidity (for safety and predictability) and fluidity (for actually handling the messiness of the real world).
Right now, most agents seem to fall into two traps: they are either so rigidly scripted that they break the moment a user deviates from the happy path, or they are so fluid that they hallucinate and drift off-task.
This post argues that we need "Structural Plasticity" in our agent architectures—allowing the system to dynamically re-route its own logic flows when it hits a blocker, rather than just forcing a retry or failing silently. It's an attempt to define a middle ground where an agent can be adaptive enough to be useful, but structured enough to be safe.
Would be curious to hear how others are handling exception handling and "adaptation" in production agents right now.