I've been obsessed with finding the perfect entry point for building something useful with this AI agent stuff. Not the flashy demo crap - something real people would actually pay for.

Here's my take: start super narrow. Cyber security competitive analysis.

Why? Because:

• With each tech advancement, cybersec only expands further.
• The tools for market research are expensive as hell but often mediocre.
• The analysis workflow is tedious and repetitive.
• I have enough technical literacy in this space to evaluate solutions without needing to be a complete domain expert.

The problem with current tools

Market analysts spend hours gathering intel on competitors' offerings, parsing technical docs, and comparing feature sets. It's the perfect task for automation.

I just checked Crunchbase for competitors of Adaptiva (endpoint management) and it gave me Tenant (property management software) and Goliath Technologies. How is property management related to endpoint security patching? The tools are connecting dots that don't exist. Even when there's a slight connection (like Goliath monitoring virtual desktops), it's missing the forest for the trees. They're solving completely different problems!

The opportunity

This is the perfect opportunity. The data exists, but the intelligence doesn't. Companies are paying thousands for tools that give them nonsense they have to manually filter. Imagine a system that could actually understand what companies DO, not just pattern match keywords. That's the gap.

So I'm picking a list of 600 companies in security to collect data on them, to build an agentic flow that will understand what they do and build competitve links between them. Not some garbage "they both mention cloud" connections - actual competitive analysis.

The approach

The trick? Breaking it into tiny tasks that chain together. Each task gets its own context, its own goal, and passes data to the next one. First task: grab the company description from their website and latest press releases. Second: extract their actual technical capabilities (not marketing fluff). Third: determine which market segments they ACTUALLY serve.

I'm not trying to boil the ocean here. That's where these projects fail. Just a modest pipeline that gets BETTER at understanding security products than humans manually reviewing spreadsheets.

The advantage

The beauty is that I don't need to hardcode knowledge about every security domain. The system learns as it processes. Each new company improves the map. Will it be perfect? Hell no. But it'll be 10x better than analysts spending days on manual research only to get it wrong anyway because nobody can keep up with 600+ companies.