Bittensor is an incentive network. Many subnets compete on model quality. They optimise for benchmark scores. That phase is necessary. It is not sufficient. The long-term value in decentralised AI does not come from isolated models. It comes from networks that solve real-world economic problems. @manakoai is building production-grade visual infrastructure. That changes the role of the subnet. Instead of training a single general-purpose vision model, the subnet becomes a distributed engine for continuously optimising production elements. Production vision is not static. Latency requirements change. Hardware profiles change. Vertical constraints change. That creates a constant optimisation surface. A centralised team can improve models incrementally. A decentralised incentive layer can improve them continuously. That’s the opportunity. The subnet is not a research playground. It is a competitive model factory. Miners compete to produce better production-ready elements. Validators score based on deployment-relevant metrics — not just academic benchmarks. The result is compounding optimisation. As Manako scales deployment across enterprises, it generates real performance feedback. That feedback informs scoring. Scoring drives incentives. Incentives drive better elements. The subnet becomes economically aligned with real-world adoption. That is our difference. We are aligning emissions with infrastructure value creation. When Manako becomes the standard execution layer for production vision, the subnet becomes the optimisation engine underneath it. That creates structural demand for better elements. Better elements improve endpoints. Better endpoints increase adoption. Adoption increases economic weight. The loop compounds. Far beyond speculative value. It is usage-aligned incentive design. Infrastructure markets reward standards. Standards reward optimisation. Optimisation is what decentralised networks are structurally good at. The subnet is not a side experiment. It is a force multiplier. Centralised orchestration. Decentralised optimisation. That combination is difficult to replicate. As Manako grows, the subnet benefits from: - Increased demand for production-grade elements - Continuous benchmarking against real deployments - Economic weight tied to utility, not hype In the long term, the subnet becomes the most efficient place to produce production-ready computer vision components. That is how decentralised networks create durable value. Manako defines the standard. The subnet accelerates the optimisation. Together, they compound.

Video has unique challenges which force it to operate on the edge in production. So we turned the subnet into a vision distillation machine, and the numbers are now clear.

I truly enjoyed telling this story as a way to demonstrate were building at score and manako

Returning from forwarded engineering @manakoai in Lyon with @MaxScore and @arnod3f Another subnet 44 vision skill, created by bittensor miners on @webuildscore , deployed and running 24/7 on a mini pc in a petrol station permissionless intelligence -> trusted operator -> client This is the pattern which brings @bittensor to the world. What time to be alive

. @MaxScore on stage at Proof of Talk talking enterprise adoption of Bittensor. No one better. King

Proof of Talk @proofoftalk was the perfect place to bring the world of @bittensor, finance and crypto up to date with what we’re building on subnet 44 @webuildscore provides an open marketplace for deployable vision capabilities @manakoai is an enterprise product combining those vision capabilities into vision agents, running locally where the cameras are at.

Today’s announcement is a milestone for our ecosystem. To match it, we’re locking $1M in perpetual conviction today. We’re here for the long game.

Bringing vision intelligence to the world

Honoured to be part of this excellent event. And excited to talk about what we’re building with @webuildscore, @manakoai on Subnet 44. The real work is making decentralised vision systems useful, measurable, and production-ready. Looking forward to Paris.

building products has changed. I started using AI for code ~2 years ago. autocomplete. marginal speedup. fine. late 2025 the harnesses caught up to the models. claude code, then codex. a task that took days started taking hours. then the cycle just… collapsed. now an agent finishes something and looks at me …. what next boss if I hesitate, it feels like time is bleeding. if I’m vague, it confidently builds the wrong thing, beautifully, in an afternoon. so the bottleneck isn’t engineers anymore. it’s taste and clarity. knowing what’s worth building, and compressing a month of thinking into the next instruction fast enough to keep up. smart engineers are multipliers on this. (we’re hiring, hmu) I’m basically a chief unblocker now. context-holder. direction-giver. the human in the loop is mostly there to keep the loop pointed somewhere worth going. agents will probably absorb this part too eventually. but for genuinely novel stuff I think a human still has to hold the thread. and honestly it’s the most fun I’ve had building in years. golden age. keep a hand on the agents. touch grass.

We’re building the vision intelligence substrate - and making it available to anyone and everyone

Mine bittensor

We’re hiring a senior platform/backend engineer at Manako @manakoai We’re building vision systems that have to work outside the lab — where deployment, runtime reliability, workflows, and product/backend integration actually matter. If you’ve built strong systems in infra, orchestration, edge, or developer platforms, and want to help turn model capability into real production systems, I’d love to talk. DM me if that’s you, or if someone comes to mind.

It was excellent to watch Greg present this in Spain last week. building deep in the stack as a subnet owner, threads like this are a valuable reminder of the broader stakes and validation that we're working on exactly the right things

Fantastic to spend time with many of the leaders in the Bittensor $TAO community at Yuma's first ever summit. The vibes, enthusiasm and camaraderie reminded me of DCG's first Summit in 2015 x.com/BarrySilbert/status/20… Truly an inspirational group. Big things are happening

The next major enterprise AI layer won’t live only in text boxes and copilots. It will live in the physical world. Cameras, sensors, operations, environments — turned into systems that can understand what’s happening and trigger action in real time. That’s why our alliance with @PwC_France matters. We’re building toward deployable physical AI for enterprise at real scale — not PoCs that look good and go nowhere. Huge moment for Manako.

AI is already doing real work. Codex. Claude Code. They don’t just suggest—they ship. Quietly removing hours of human effort. That’s the leading edge. But it points to a bigger shift Jensen Huang recently called out: from reasoning to work. And once you see it, the pattern is clear: the winners aren’t single models, they’re systems — composed, orchestrated, repeatable the bottleneck isn’t intelligence, it’s reliability, evaluation, and trust and more of this work moves closer to the source — edge, on-device, in situ Coding is just the first place this becomes undeniable. Now zoom out. Most domains don’t have tight feedback loops. No clear benchmarks. No incentive to keep improving once something “works.” That’s where Bittensor fits. It turns performance into a game you can’t fake. Specialised models, measured on real tasks, competing continuously. Better outputs, more reward. Worse outputs, you disappear. Not one frontier model. A network that compounds capability, domain by domain. The gap now isn’t whether AI can do the work. It’s how fast you can make it reliable. Bittensor accelerates that.

This week we shipped workspace-scoped model management on @manakoai, along with a shared library flow. Approved models can now be published once, then copied or forked into customer workspaces with lineage, auditability, and role-based controls intact. We also tightened the subnet and benchmarking tooling behind the scenes to make training, validation, and deployment loops more reliable.

We’re live on @twistartups @MaxScore is speaking on the future of vision AI and Score’s & @manakoai approach to programmable vision.

The @manakoai desktop agent is born... No cloud pipelines. No egress tax. No latency games. Download → it finds your cameras → deploy vision agents locally → inference runs on your hardware. Only evidence telemetry hit the cloud. We're moving vision compute to where it belongs: the edge.