Ibrahim Muhammad

The three places value moves next: → Physical world (robotics) → Proprietary data → Deep domain knowledge Full breakdown: ibrahimmuhammad.com/posts/trends-s…

The meta-pattern: → Voice agents sound human → Software becomes a commodity → MCP makes integration cheap → Context is the new bottleneck

Agents are eating software. 15 shifts shaping tech in 2026. A thread 🧵

Over the last year: "Claude, make this skill or MCP server for me."

When trying to build an initial proof of concept. Before ChatGPT: "Can I prototype this in a Jupyter notebook?" After ChatGPT: "ChatGPT, can you write this script for me?" After Claude Code: "Claude, make this app for me."

But the top layer is changing. The agent starts to look like a runtime. Natural language starts to look like a programming interface. And skills start to look a lot like apps, with directories to find them and package managers to install them.

You still need code for the underlying systems: integrations, infrastructure, permissions, reliability, and all the deterministic parts that have to work. You still need models for reasoning, language, perception, and generation.

A new computing paradigm is emerging where the agent is the runtime, natural language is the interface, and skills are starting to look a lot like apps. Software 1.0 was hand-written code. We told the computer exactly what to do.

@tibo_maker composekit.io

drop your website below I'll reply with 1 free tool you should build to get SEO traffic not a generic content an actual tool people would search for, use, and maybe share first arrived, first served go 👇

Plug in skills, CLI scripts, or MCP servers to connect with your existing workflows. Try it: composekit.io. Looking for the first founders to shape it.

ComposeKit: AI-native distribution for builders with tight human oversight. Built for founders who would rather be shipping code than hand-crafting LinkedIn posts. Each step has a configurable prompt and can be done by you, an agent, or both.

This is definitely an interesting idea and I'm curious to see how it evolves with stronger models. For now though, it's not making it into my stack. But with 53k+ stars on Github, am I missing something?

In practice though, it fell short for me. For software tasks I prefer the tighter feedback loop that Claude and Codex provide. Token waste was another problem, a heartbeat periodically wakes agents to check for work even when there's nothing to do.

I tried to create a team of agents to build software for me. Here is what I learned. I wanted a single place to coordinate all my agents. Paperclip lets you create agents with roles like CEO, CTO, and Software Engineer, then assign them work to run your company.

OpenClaw has been getting a lot of attention lately, and I think skills are the underlying reason. Not the agent itself. The fact that anyone can extend it, build their own tools, and share them. What skills have you been finding useful beyond coding?

I use many other skills too, ones I installed from a directory, much like installing an app. The difference is I can personalize them just by talking to an agent. No SaaS subscriptions. You pay for the LLM tokens you use.

I built three tools last month that only I will ever use. One plans my day. One tracks my reflections. One filters content so I stop doom scrolling. None of them are apps. They're skills, instructions and scripts that store data in markdown files.