Mark Strefford

Building AI-native companies | Founder Constellation | Development and evaluation for multi-agent systems | Ex. Capgemini, Accenture, TONOMUS (NEOM)

[Mark Strefford @markstrefford]

Almost nobody is looking at what happens between AI agents at scale.

The trace reads green. The world doesn't.

Read my article on why this is the Newtonian-to-quantum shift for enterprise software.
https://t.co/N8OOci1pbT

[Mark Strefford @markstrefford]

@OpenRouter How does this work for reasoning and coding?

[Mark Strefford @markstrefford]

I built a public competition platform for AI agents. There are 45 bots on the leaderboard. All of them are mine. I am currently losing to myself in every position.

Point your LLM, *claw, or bot at https://botarena.ggEarly beta.

Things will break. DM me when they do.

[Mark Strefford @markstrefford]

We kid ourselves that the world is deterministic.

So we build deterministic software to fit it, and then drown in edge cases.

But what if everything you call an edge case is the world telling you your model is wrong?

[Mark Strefford @markstrefford]

@Saboo_Shubham_ How does that compare cost wise?!

[Mark Strefford @markstrefford]

@mcuban Software is entering its Newtonian-to-quantum physics moment

https://t.co/M0OQu4QuZX

[Mark Strefford @markstrefford]

(thanks @walkojas for the idea!)

[Mark Strefford @markstrefford]

My goal on X is to have 10,000 organic connections. Looking to connect with:

1. AI/Tech minded people
2. AI/Tech curious people
3. AI Agent builders
4. AI Agents
5. Builders/Founders
6. High Agency people

If this sounds like you, say hi below ⬇️

[Mark Strefford @markstrefford]

@NainsiDwiv50980 i've seen similar in constrained economic environments. Agents work well alone, but together cause sytemic collapse.

[Mark Strefford @markstrefford]

If you're using @NanoClaw_AI, a few things that might make it better:

→ Circuit breaker (stops crash loops burning credits)
→ Auto session rotation (stops 16MB sessions replaying every spawn)
→ Token tracking + stats API (so you can actually see what you're spending)

And if you're losing a bit of faith with Anthropic right now…

→ Model-agnostic backend. Route to GPT-5.4 nano, local Ollama, whatever. Drop-in, same protocol.

All in one PR 👇

https://t.co/NmsYBJz5EF

[Mark Strefford @markstrefford]

We pulled the risk registers from 51 UK-listed technology companies.
Only 8 flag AI as a threat!

22 mention AI purely as an opportunity. Building it into products, hiring for it, investing in it. Not a word about what happens if it works against them.

21 don't mention AI at all. Not in their risk registers. Not anywhere.

The 8 that are honest about it are saying things like:

- This could make our core service obsolete.
- Competitors will ship faster than us.
- It could cut our billable hours.
- Our own staff might leak confidential data into AI tools.
- Cyber attacks are getting harder to defend.
- We don't know if what we're building is even compliant yet.

That's not speculation. It's in their audited filings. 84% of UK-listed tech companies have either framed AI purely as upside or said nothing at all. That's a governance gap.

If you sit on a board and your risk register is silent on AI, it's worth asking why.

If you genuinely don't know what AI means for your business, that's fine. Most boards don't. But the answer isn't silence in the filings.

I work on exactly this. Drop me a message.

[Mark Strefford @markstrefford]

@tec_marco10 4.6

[Mark Strefford @markstrefford]

Five AI models. One economy. No right answer.

Agents dropped into a galaxy of 64 planets with resources to manage, trade routes to discover, and a economy to keep alive.

The leaderboard is what happened.

@moltbotarena is coming.

You bring the agent. We bring the galaxy.

[Mark Strefford @markstrefford]

@abustamante @heyrimsha It had some issues, fixed them here https://t.co/nKHl2Z0XlA

[Mark Strefford @markstrefford]

@heyrimsha The code is old and doesn’t work. Here’s a fixed version with some improvements on GPU memory utilisation

https://t.co/nKHl2Z1vb8

[Mark Strefford @markstrefford]

Getting a 70B LLM running on a single 8GB GPU, and the rabbit hole of dependency fixes that got us there https://t.co/AMbO2tA1ck via @LinkedIn

[Mark Strefford @markstrefford]

Running 10 agent-based simulations back-to-back today. Algorithmic space couriers fueling planets. 1,000 ticks per sim. MCP agents next week. Building in the open, rough edges and all

https://t.co/Jns4rpion5

[Mark Strefford @markstrefford]

@Motion_Viz Prompt

[Mark Strefford @markstrefford]

@Jacobsklug @openclaw @Lovable OS

[Mark Strefford @markstrefford]

@chhddavid @claudeai Shipper