this past week, anthropic released mcp apps, which lets tools like figma, jira, and amplitude render familiar, consistent, interactive ui components directly inside claude chat.
this is a pretty direct response to openai’s apps sdk launched last october, which lets developers (like spotify) build apps that run natively inside chatgpt.
in my opinion, this could be the most important ux trend for product managers to watch right now: more user workflows beginning and ending within ai chatbots.
here’s how this might change our lives as product managers:
imagine your cpo pings you: “why are signups tanking?”
you open chatgpt and ask it to investigate. your company has already connected chatgpt to sentry, zendesk, and snowflake.
chatgpt spins up three agents, while you go make a coffee:
one agent checks sentry for downtime or error spikes
one agent scans zendesk tickets for user-reported issues
one agent queries snowflake data for anomalies and segment-level drops
you come back and review the results, rendered within chatgpt using native ui from sentry, zendesk, and snowflake (courtesy of mcp apps / apps sdk).
nothing unusual in sentry or zendesk, but the data shows a sharp signup drop among users on a new chrome version (26.2). everything else looks normal.
from the same chat, you ask chatgpt to:
create a jira ticket for engineering to investigate
draft a notion summary of the bigquery findings
finally, you open notion, make a few edits, then share the polished doc with your cpo.
notably, the above workflow happened almost entirely within chatgpt, and mostly autonomously. historically, this same workflow would have happened mostly in individual saas tools, mostly by users themselves.
this is the future i believe pm’s and founders need to begin designing for:
chat as the primary user surface
users descend into native tools only for “finishing touches”
agents drive most work; humans mostly review
how will your product perform when most user workflows are driven by agents, and occur entirely within ai chatbots?