We just claimed our .agent domain and joined Agent Community! Get yours now and help shape the future of autonomous agents https://t.co/ew5mQcCPsm @agentcommunity_
I’m increasingly convinced most custom software is just expensive glue.
Integrations, adapters, CRUD, orchestration, internal dashboards, brittle workflows.
In the AI era, a lot of that should move from coded software to configured systems:
- reusable tools
- bounded agents
- explicit contracts
- human escalation where needed
Compose, don’t rebuild.
Adapt, don’t maintain.
The teams that understand this early are going to look unfairly fast.
Most “AI products” are still regular software with an LLM bolted on top.
That’s not agentic architecture. That’s garnish.
The real shift is:
- agent as kernel
- MCP tools as capability layer
- APIs / DBs / PDFs / websites / legacy systems as raw material
Once you model the stack that way, you stop asking “where do we add AI?” and start asking “what parts of this system should still be hardcoded at all?”
That’s a much more interesting question.
Ghostty is leaving GitHub. I'm GitHub user 1299, joined Feb 2008. I've visited GitHub almost every single day for over 18 years. It's never been a question for me where I'd put my projects: always GitHub. I'm super sad to say this, but its time to go. https://t.co/DQDemHdytV
The market rewards AI theater.
Reality rewards operational sharpness.
I’m much more interested in founders who can answer:
- where does the data come from?
- what gets escalated?
- what stays deterministic?
- what breaks when the source changes?
That’s the difference between a demo and a company.
AI in travel gets real very fast.
You start with “build a hotel seller.”
Then reality shows up:
- room types named differently across brands
- occupancy rules that don’t map cleanly
- partial pricing
- external booking engines
- inconsistent data everywhere
If your architecture only works on polished demo inventory, it doesn’t work.
I genuinely cannot hear the word "align" anymore. Everything and everyone have to "align" to everything and everyone. Corporate is doomed, let's align on that 😂
What makes MCP interesting is not function calling.
It’s the architectural consequence of interchangeable capabilities.
Swap the booking backend.
Keep the kernel.
Swap the data source.
Keep the behavior.
Swap the surface.
Keep the system.
That’s a much bigger shift than “LLMs can use tools now.”
Hot take: most custom business software is still just expensive translation work between systems.
That’s why I’m so bullish on composable tool layers.
If an agent can interpret semantics and tools provide stable capability boundaries, a lot of the old application layer starts to look unnecessary.
That’s the opportunity: not better glue. Less glue.
The hard part of agentic software is not the agent.
It’s everything around it:
- permissions
- tool boundaries
- escalation rules
- memory
- versioning
- auditability
If your whole product story is “we have a smart model,” you’re still at the toy layer.