Ep. 107 - Dylan Uys & Ken Miyachi
Dylan @_dubm and Ken @kenjon are building @yugosai
Timestamps
01:00 Dylan Uys: From BitMind to Yugos
04:09 AI coding workflows and model choice
06:33 The harness versus the model
11:31 Balancing intelligence, cost and speed
13:21 Keeping the model and provider catalog current
16:43 Routing across Bittensor services
20:47 Routing versus multi-agent orchestration
26:01 Benchmark gaming and validation
30:03 Can a routing system build a moat?
31:44 The value of real-world traffic
33:37 Yugos roadmap and custom provider choices
35:55 Local models and privacy
38:24 Agents in the research workflow
40:24 Ken Miyachi
42:29 The future of agent harnesses
47:02 Building a business around the router
52:58 The human role in software development
55:24 Distribution and lessons from BitMind
58:35 Competing with well-funded AI companies
1:00:56 Vendor lock-in and changing models
1:04:01 Yugos go-to-market plans
1:04:57 Token spending and the value of AI work
1:08:51 Research versus production development
1:10:59 Training routers for different user needs
1:16:18 Consumer AI and agent business models
1:21:01 Provider incentives and customer trust
1:22:44 Why distribution matters for Yugos
1:25:12 Open source, monetization and defensibility
1:28:26 Bittensor development speed and security
Can agents use experimental feedback to establish new material capabilities that hardware teams can reliably design around?
Today weโre releasing SDL-2, an open evaluation of how agents choose experiments and decide whether a material process will reliably meet a requirement.
Weโre interested in the full process of scientific investigation. That means discovering which relationships matter, choosing experiments that distinguish competing explanations, revising an explanation when results surprise and knowing when the evidence is strong enough to act.
The experiments run in a digital twin of magnet processing, calibrated on measurements from our campaign with Rensselaer Polytechnic Institute (@rpi). Agents investigate unfamiliar powders with an uncertain physics model, a limited experimental budget and results that vary from batch to batch.
Across 32 simulated worlds, frontier models reached correct decisions supported by their own measurements in 44โ50% of investigations, compared with 25โ28% for other language models and classical experiment design. One in three of their correct answers were unsupported.
Our larger ambition is for materials development to keep pace with hardware design, so teams can design around the capabilities they request, not only those that already exist. SDL-2 makes one part of that measurable, how agents investigate.
We're releasing the evaluation code, harness, digital twin and all 256 investigation traces.
Technical Report: https://t.co/zrSPF2Thll
Code: https://t.co/GDQAYbGTan
Dataset: https://t.co/Jz4DTeLGTr
Yugos is an system one decision model to optimize inference choices. Model, provider, effort.
Specialized decision models are the future. Low latency with massive speed and cost improvements over AI reasoning.
Our team has been switching models and providers consistently over the past year as we tried to balance token cost and intelligence. Cursor, OpenAI, Claude, Open source models. Always testing new model drops. Now this process is a bittensor competition where a single endpoint always serves the best model for your task
The last AI model you'll ever choose.
New models launch every week. Pricing changes. Intelligence improves. Providers compete on speed, cost, and performance.
Keeping up with which model to use, from which provider, and at what reasoning effort is becoming a full-time job.
Introducing Yugos.
One intelligent decision layer for AI inference that continuously optimizes for intelligence, cost, and speed.
Yugos automatically selects the best model, provider, and reasoning effort for every request.
As new models launch, pricing changes, and performance evolves, Yugos adapts automatically.
All with full transparency into model selection, usage, and costs.
Stop choosing models. Let Yugos optimize intelligence for you.
Live today. โ
@Yizhiyangfund@const_reborn a lot of token designs can be better than stable coins. diem is an interesting example here. think this is a really good idea tbh. wait for details to come out before reaching judgement ๐ซก
Here is @const_reborn's full keynote at @ExploitSummit
It is worth the 36 minutes, but I'll give you the cliff notes.
> Affine 1, a model trained by anonymous miners entirely on Bittensor, matches the intelligence score of Anthropic's Opus 4.1.
> 23 of 128 subnets now have paying customers
> Gamma tokens are coming to turn subnet usage credits into something agents can buy from any chain.
> "We're not afraid of hyper-intelligent bot swarms on the internet. We use them to optimize our technology."
We're excited to announce that we're going to be joining Y Combinator in Fall 2026.
The team is super pumped to be working with @golda and co to continue on our journey of creating the best in class open source models for agentic commerce.
Announcing Teutonic-II-110B: a frontier permissionless training of a 110B LLM
We are scaling 10x beyond our previous model and 50% larger than any prior decentralized run.
You can't trust your own eyes anymore. AI-generated faces, voices, and video keep getting harder to catch - and cheaper to make.
With synthetic media is scaling faster than anyone can moderate it, detection has to scale with it.
That's why @kenjon, Co-founder & CEO of @bitmind, is building the detection layer that fights back - on #Bittensor.
Before BitMind, Ken was a software engineer and technical lead at organisations including @NEARFoundation and @amazon. Now he's building the infrastructure we need in the next gen of AI.
He's on the Exploit stage next month: https://t.co/CgJmJoRXKk
Congrats @ai!
Canonical has been invaluable to @bitmind from day one. First believers, opened so many doors, and the brainstorming/advice/feedback has been deeply impactful
When I started Canonical, the mission was simple: back technical founders building core mission-critical infrastructure before it's obvious.
Today, I'm extremely proud to share that Canonical has been selected for the 2026 @raiseconference Select cohort alongside an incredible group of emerging fund managers.
To our portfolio founders, LPs, and community partners - thank you for trusting us.
Looking forward to connecting with LPs & GPs at The Presidio in SF on Oct 22, 2026. Drop a DM if you'll be there!
Today, weโre officially launching OpenRoboto โ Bittensor Subnet 80.
Weโre sharing the full vision: open weights, open data and open benchmarks, coordinated through Bittensor to continuously improve robotics models.
.@bitmind just published a paper placing BitMind Forensics among the worldโs top deepfake detection systems.
It matched the best commercial image detector, beat the best commercial video detector, and outperformed top models on major benchmarks.
Hereโs the Bittensor edge:
'How fast should the protocol change?'
@askyoav: the protocol's moving too fast.
@0xcarro: itโs not fast enough.
Bittensor must move fast to compete with centralized AI - but constant change is bad for businesses building on it.
See the debate on stage at Exploit:
https://t.co/CgJmJoRXKk
For my first post, Iโm sharing a letter @NVIDIA signed on why open models matter.
AI will transform every industry, power every company, and be built by every country.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.
The world needs both frontier closed models and frontier open models.
https://t.co/AUKzoQ5Ikb