The next AI SaaS opportunities might not be AI apps — they might be AI infrastructure
Title: The next AI SaaS opportunities might not be AI apps — they might be AI infrastructure
I've been tracking emerging AI products, GitHub projects, and market signals to understand where new startup opportunities are appearing.
My biggest takeaway:
The AI market is moving from "building AI features" to "building infrastructure that makes AI agents actually useful."
A lot of AI apps are becoming commoditized.
The interesting opportunities seem to be around solving the problems that appear when people use AI seriously.
1. Agent Memory & Context Persistence
This is the strongest signal I've found.
The problem:
AI coding agents still forget.
Developers using tools like Claude Code, Cursor, and Codex often need to:
- explain the project again
- restore previous decisions
- manually maintain context files
- repeat previous instructions
The opportunity:
A memory layer for AI agents:
- persistent project knowledge
- cross-agent memory
- team-shared context
- better retrieval and compression
The question:
Will memory become a core feature inside existing AI platforms, or will there be room for independent companies?
Other areas I'm watching:
2. AI Agent Cost Tracking
As teams run more agents, understanding:
- token usage
- cost
- performance
- ROI
will become increasingly important.
3. AI Agent Security
More autonomous agents create new problems:
- permissions
- sandboxing
- data protection
- compliance
4. Vertical AI Workflows
The strongest businesses may still come from focused industries:
- healthcare
- recruiting
- legal
- operations
because customers pay for outcomes, not technology.
My current belief:
The next generation of AI startups may not win by creating another chatbot.
They may win by making AI agents:
- remember
- reason over company knowledge
- operate safely
- integrate into real workflows
I'm curious:
For founders building AI products, what problem feels the most painful right now?
- AI agents forgetting context
- Getting reliable outputs
- Finding customers
- High AI costs
- Something else?
The memory point is interesting, especially because “developers complain about this” and “there’s room for an independent company here” are two different signals.
What have you seen that makes agent memory look like a standalone buying problem rather than something users expect Claude, Cursor, or Codex to eventually absorb?
Great question. I think the key distinction is between a feature and a platform layer.
Claude, Cursor, and Codex will likely improve their own memory systems, but independent opportunities may exist around cross-model memory, user-owned context, and persistent knowledge across different AI tools.
For example, a developer may use Cursor for coding, Claude for research, and other agents for workflows. The question becomes: who owns and manages the long-term memory layer across these experiences?
Of course, the challenge is whether this becomes a must-have workflow or just a nice-to-have feature. That’s the part I’m still watching.
I appreciate you taking the time to explain your thinking.
I'd be interested in continuing the conversation by email if you're open to it. What's the best email to reach you on?
Thanks for your interest!
My email is: zhixingheyi2023@gmail.com
I’d be happy to continue the conversation there.
I’m currently exploring opportunities around AI products, infrastructure, and market signals, and I’d love to hear your thoughts as well.
Thanks! I’ve just sent it over.
Looking forward to hearing your thoughts whenever you have a chance.
Thanks, I’ll take a look and share my thoughts. Appreciate you sending it over!