What I love most is how you reframe “agents” from a mystical new category into behaviors people are already using in tools like Cursor, Claude, Perplexity, and Notion AI, which makes the whole space feel actionable instead of abstract.
This demystification is exactly what the market needs right now. Everyone's throwing around "agentic AI" but half the time they're just describing a chatbot with a few if-statements.
Your three universal types (information, interaction, operational) map perfectly to what I've been seeing in the competitive landscape. The winners aren't the ones with the most sophisticated AI—they're the ones who picked the right agent type for their actual problem.
The confusing part is that the infrastructure spend is accelerating ($175B+ from Alphabet alone) while software stocks tanked $285B in one quarter. That's not market noise—that's a bet that value is moving to orchestration layers. Hyperscalers win, middleware gets compressed, SaaS either integrates or dies.
I wrote up the Feb 2026 market data and the pattern is basically cloud consolidation all over again (https://thoughts.jock.pl/p/ai-agent-landscape-feb-2026-data). Five years from now we'll look back and say "obviously" but right now everyone's still pretending their point solution has moat.
Ah I went for Claude code + obsidian for the content system … to keep easy access to underlying data. But it’s limited when it comes to tracking analytics, conversations etc… are you happy with your supabase system ?
Definitely, for structured data, I like keeping it in Supabase. One of the best use cases is that Supabase has very good MCP support, so I can just ask questions in Cursor directly and Cursor can help me run sql queries to fetch data.
Same as you, I naturally distinguished agents from other activities until I tried to connect the dots across the whole story to keep myself from feeling overwhelmed.
I’m still wowed by how capable AI agents can be, but I don’t feel disconnected anymore.
Thanks so much for this thoughtful overview AI agents, which clearly separates the agentic hype from the useful tools! 🙏 MCP looks especially promising to me and building an MCP server has been on my to-do list for a while.
Very comprehensive, thank you 🙏🏾
Glad it resonates :)
God that is thorough... super enjoyable
Thank you so much, Paul :)
What I love most is how you reframe “agents” from a mystical new category into behaviors people are already using in tools like Cursor, Claude, Perplexity, and Notion AI, which makes the whole space feel actionable instead of abstract.
Glad you loved this Alex :)
This demystification is exactly what the market needs right now. Everyone's throwing around "agentic AI" but half the time they're just describing a chatbot with a few if-statements.
Your three universal types (information, interaction, operational) map perfectly to what I've been seeing in the competitive landscape. The winners aren't the ones with the most sophisticated AI—they're the ones who picked the right agent type for their actual problem.
The confusing part is that the infrastructure spend is accelerating ($175B+ from Alphabet alone) while software stocks tanked $285B in one quarter. That's not market noise—that's a bet that value is moving to orchestration layers. Hyperscalers win, middleware gets compressed, SaaS either integrates or dies.
I wrote up the Feb 2026 market data and the pattern is basically cloud consolidation all over again (https://thoughts.jock.pl/p/ai-agent-landscape-feb-2026-data). Five years from now we'll look back and say "obviously" but right now everyone's still pretending their point solution has moat.
Very well said! At least in my hope, a few years from now, the agentic AI concepts and everything around it be obvious for most people.
I think so. For now - it is still a “nerd domain” :D
This is incredibly helpful, Jenny, thank you!
Now that I'm in Cursor most days, I'm going to try my hand at MCPs. Thanks for that post, too, lol.
My pleasure Kim! I'm so glad they become useful for you :)
Thanks Jenny for adding to PM perspective article. This article is so more in detailed for everyone. you nailed it :)
Thank you Poojitha!
Ah I went for Claude code + obsidian for the content system … to keep easy access to underlying data. But it’s limited when it comes to tracking analytics, conversations etc… are you happy with your supabase system ?
Definitely, for structured data, I like keeping it in Supabase. One of the best use cases is that Supabase has very good MCP support, so I can just ask questions in Cursor directly and Cursor can help me run sql queries to fetch data.
Nice breakdown as always! Llet's not forget the most common agent type: "the AI workflow that people call AI Agent" 😉
Oh haha yes, the AI workflow too!
Thank you, I was thinking maybe I was missing a fundamental part of AI - now I know I have been using agents heavily all along
Same as you, I naturally distinguished agents from other activities until I tried to connect the dots across the whole story to keep myself from feeling overwhelmed.
I’m still wowed by how capable AI agents can be, but I don’t feel disconnected anymore.
Thanks for reading, David.
Thanks so much for this thoughtful overview AI agents, which clearly separates the agentic hype from the useful tools! 🙏 MCP looks especially promising to me and building an MCP server has been on my to-do list for a while.
Absolutely, building an MCP is actually a lot easier than you'd expect :)