I have to say that I haven't seen tech articles with a good level of clarity and sensible prediction for a while (I have been swimming in the AI content farm and Linkedin shit posts for a while). Thank you.
I've been thinking a lot about MCPs in the context of marketplaces and how this will play out. I'm intrigued to hear any thoughts on discoverability and how marketplaces might play a part here.
Arguments on vertical focus and redistribution on value are pretty on point.
+1 on the fragmentation dangers facing MCPs.
I think there is also the underlying risk of how well the "Agents as the everything interface" assumption will hold. If it turns out that triaging everything too an LLM call is too expensive or too slow in more scenarios than expected, we might not get the scale at which many of these value capture arguments hold.
Great analysis, thanks for sharing! The parallels are undeniable.
IMO MCP - and any other protocol that comes out - is not an API 2.0 like GraphQL. GenAI changes the way users interact with software and the underlying data. That is the drive behind IDEs like Cursor becoming creativity expression tools for the average person. The classic use cases of the web evolve which might have an impact on value accumulation. Would love to see an analysis about that!
Thank you for sharing. This piece is well written, insightful and compelling. Lots of food for thought on how to differentiate in the market. I notice that n8n is not in your horizontal/vertical mcp players list - this seems an oversight as it looks like a well entrenched player combining infrastructure, protocol and abstraction over langchain
How can you build a vertical MCP product by just owning the server? Seems like any product sufficiently vertical needs to own both the server and client?
Enabling MCP consistently and in simple manner still needs to be solved. Question remains whether the ‘cloud provider’ infrastructure layer handles it or the SSO vendors tackle it.
as you noted, it is early innings and would agree that a standalone commercial approach will not be viable.
It might still be a good idea to build in the MCP infra space with the goal of getting acquired by a cloud player! Startups can move fast and get ahead of big tech. But in these horizontal domains, ultimately, the winners will be big tech, sadly.
Great post
Thank you!
Excellent read! Thank you
Once again, superb thinking. I always look forward to your contrarian insights 👏
I have to say that I haven't seen tech articles with a good level of clarity and sensible prediction for a while (I have been swimming in the AI content farm and Linkedin shit posts for a while). Thank you.
Thank u for this excellent deep research. Vital lessons here = Playbook shifting if we are all smart enuf
A wonderful write.
I've been thinking a lot about MCPs in the context of marketplaces and how this will play out. I'm intrigued to hear any thoughts on discoverability and how marketplaces might play a part here.
Lots of good ideas here.
Arguments on vertical focus and redistribution on value are pretty on point.
+1 on the fragmentation dangers facing MCPs.
I think there is also the underlying risk of how well the "Agents as the everything interface" assumption will hold. If it turns out that triaging everything too an LLM call is too expensive or too slow in more scenarios than expected, we might not get the scale at which many of these value capture arguments hold.
Great analysis, thanks for sharing! The parallels are undeniable.
IMO MCP - and any other protocol that comes out - is not an API 2.0 like GraphQL. GenAI changes the way users interact with software and the underlying data. That is the drive behind IDEs like Cursor becoming creativity expression tools for the average person. The classic use cases of the web evolve which might have an impact on value accumulation. Would love to see an analysis about that!
A fantastic piece of work!
Great overview of MCP and the evolvement of it
Thank you for sharing. This piece is well written, insightful and compelling. Lots of food for thought on how to differentiate in the market. I notice that n8n is not in your horizontal/vertical mcp players list - this seems an oversight as it looks like a well entrenched player combining infrastructure, protocol and abstraction over langchain
How can you build a vertical MCP product by just owning the server? Seems like any product sufficiently vertical needs to own both the server and client?
Yes, the illustration is basically to show which models people use to build the clients!
Enabling MCP consistently and in simple manner still needs to be solved. Question remains whether the ‘cloud provider’ infrastructure layer handles it or the SSO vendors tackle it.
as you noted, it is early innings and would agree that a standalone commercial approach will not be viable.
It might still be a good idea to build in the MCP infra space with the goal of getting acquired by a cloud player! Startups can move fast and get ahead of big tech. But in these horizontal domains, ultimately, the winners will be big tech, sadly.
💯