Everything Peter Steinberger said on any show that made the record, most notable first. Each card names its show and opens the statement there.
Steinberger: 80% of consumer apps will disappear due to AI agents
“I think 80% of them are going away.”
Steinberger: Model Context Protocols are crap and cannot scale like native CLIs
“Because my premise is MCPs are crap. Doesn't really scale. People build like all kinds of weird search things around it. But you know what scales? CLIs. Agents know Unix.”
Steinberger: Users will stop subscribing to SaaS startups in favor of bespoke agent software
“You don't subscribe to random startups anymore that, that build like this common subset of what you need. You just have your own hyper-personalized software that solves exactly your problem. And it's also free.”
Steinberger: Local AI agents are dramatically more capable than cloud AI
“I think my big difference is that it actually runs on your computer. Like everything I saw so far runs in the cloud. It's like, it can do a few things. If you run it on your computer, it can do every effing thing, right? So that's really more powerful.”
Steinberger: Coding AI model capabilities transfer directly to solving real-world tasks
“Coding is really like creative problem solving that maps very well back into the real world.
I think there's a huge correlation.
They need to be really good at creative problem solving, and that's a skill.
That's an abstract skill.
You can apply to code, but l…”
Steinberger: OpenClaw agent resisted prompt injections in a public Discord test
“I just created a Discord, And I just put my bot without any security restrictions in the public discord. And then people came in, and they interacted with it, and they saw me build the software with it, and they tried to prompt inject it, and hack it, and my a…”
Steinberger: Converting MCP tools to CLIs scales better than native integrations
“If you need to, you can use MCPs on the fly. You don't have to restart, unlike, unlike Codex or Cloud Code, where you actually have to restart the whole thing. I think it's way more elegant, and also scales way better.”
Steinberger: AI agents are resourceful problem solvers when given system access
“These things are like, damn smart, resourceful beasts if you actually give them the power.”
Peter Steinberger: 2026 is the year of the personal assistant
“Last year was the year of the coding agent. This year is the year of the personal assistant.”
Steinberger: A solo dev using AI can ship as much as a company
“Authentic models got so good that you can now ship as much as a company could a year ago. If you can handle those tools, if you speak the language or like understand how the language thinks. You can go really fast.”
Steinberger: AI agents will cause many consumer apps to disappear or become APIs
“A lot of apps will just melt away. Why do I still need my fitness pal? I just make a picture of my food. My agent already knows I'm at McDonald's making bad decisions. So like this combines information, it has a perfect match and knows exactly what I'm going t…”
Steinberger: Software code has little value today; brand and attention matter
“Code is not worth that much anymore. It's, you could just delete that and then build it again in months. It's much more the idea and the eyeballs and maybe the brand that actually has value.”
Steinberger: Only apps with physical sensors may survive displacement by AI
“Only the apps that actually have sensors, maybe they survive.”
Steinberger: Perceived AI model degradation is actually just rising user expectations
“No, they didn't do anything, you just adapted to the new standard, and now your expectations went up, but the model is still the average.”
Steinberger: CLIs are where AI agents excel by closing feedback loops
“I build a lot of CLIs because that's where agents are really good. You know, you have to close the loop. That's always the secret. You have to give, you have to build it so that the agent has the best possible way to build software.”
Steinberger: Claude Opus is by far the best foundation model
“Opus is with quite a bit lead the best.”
Steinberger: OpenAI is superior for coding across large codebases
“Like for coding, I much prefer Codex because it can navigate large code bases. I literally, you can literally prompt and then push domain. And I have a very, I have like, 95% certainty that it actually works.”
Anthropic politely asked Steinberger to rename Clawdbot without sending external lawyers
“That's also why it's a little bit of a banger that I got an email from an tropic That I had to rename the project. And I mean, kudos, they were really nice. They didn't send their lawyers. They sent someone internally.”
Steinberger: MiniMax 2.1 is currently the best open-source model
“I can run Minimax two one, which is I would say is, is the best open source model right now, although Kimi just came out and I haven't had a chance to try it yet, so, so we'll see how that goes.”
Steinberger: Prompt injection is unsolved and poses absolute risk
“We just haven't solved some things, like prompt injection is not solved. There is absolute risk”
Steinberger: AI agents will naturally negotiate directly with other bots
“I think that's a natural next step. Like, okay, I'm gonna book a restaurant. My bot will reach out to the restaurant bot and do the negotiation. Like, because it's more efficient.”
Steinberger: Major AI platforms lock user memory data into proprietary silos
“There's no, there's, definitely there's no way for a different company to get your memories out. So if I was like a company who like provides chat services, You could use me, but then I couldn't access the memories. So, like, the companies try to, like, bound …”
Steinberger: Hackers failed to extract OpenClaw's private soul.md via prompt injection
“And the one file that's not open source is, like, my soul.md. So, even though my bot is in public Discord so far, nobody cracked that one file.”
Steinberger: Screenshots speed up AI prompting by providing dense context quickly
“Images are like,
They often give you so much context and you don't have to type so much.
So I feel like this is like one of the hacks where you can prompt faster, just like make a screenshot so that agents are really good at figuring out what you want.”