Conference news · Day 2: September 29 · Published October 1, 2026
Exploit Conference Catch Up · Day 2
Day 2 brought product demonstrations, a live funding competition and some hard questions about building on Bittensor. Here’s what the teams discussed in Montreal, with the practical takeaway from each.
The theme: move from an impressive idea to something somebody can use. Agents with spending limits, evidence behind their decisions, useful training tasks and infrastructure that can serve a customer.
Read the Day 1 catch-up →Const + TAO.com · Gamma
Proposal deep diveMake service costs easier to budget
Const and Keith Singery returned to Gamma’s pricing. The published discussion explains why businesses need predictable usage credits when TAO and alpha prices move. Const’s follow-up describes oracle-priced alpha burning; the wider proposal would allow subnet owners to spend some emissions on other subnets’ services.
The Shizzy takeThis is where Gamma gets practical. A customer should be able to budget for a job without making a token-price bet first. The next test is the final rules and an implementation people can actually use.
Carbon + Bitstarter · Pitchtensor
Competition resultPhysics AI wins the crowd’s backing
Carbon beat fellow finalist Fugal at Pitchtensor. Bitstarter reports 500 TAO raised in under 20 minutes. Carbon is building physics-AI tests intended to help engineers judge when model outputs are reliable. The announcement describes a future Bittensor subnet; a current subnet number was not established.
The Shizzy takeA live pitch makes a team explain the problem and earn attention quickly. Carbon now has backing to turn that pitch into useful work. Fundraising is the start of the execution story.
SN107 Minos
Demo · limited researcher accessAn agent runs the research loop
SN107 Minos showed an agent-driven genomic research loop on a Mac, with logs and an iPhone companion that could pause it remotely. The organizer also reports limited Studio access for approved academic researchers. The second incentive mechanism was previewed; the phone served as a control, rather than the research computer.
The Shizzy takeThat is an easier workflow to picture than another abstract AI promise: let the agent run experiments, then inspect and control the work. Research usefulness and wider availability are the milestones to follow.
SN17 404—GEN
Showcase · monetization planGames made with the tools—and a route to revenue
Ben James showcased the Game Jam entries. SN17 404—GEN’s subsequent results name The Last Bellkeeper, Farseek and SunDrift as the judged top three among 25 submissions. The organizer describes a next step of creator monetization and revenue sharing with SN17; completed sales were not reported.
The Shizzy takeFinished games give us something to look at and judge. Helping their creators earn would connect the generation tools to a repeatable business, which is the step worth watching next.
Macrocosmos + Taostats
DebateHow should emissions fund research and businesses?
Will Squires and Marcus Graichen debated research funding and commercial accountability. The organizer’s recap highlights useful outputs and timelines as questions for emissions-funded work. The session put different approaches to building on Bittensor in the same conversation.
The Shizzy takeResearch can need time, and customers can need results now. We should be able to ask what progress a team expects, how it will measure it and how long its funding can support the work.
Renno Law · Toufic Adlouni
Legal discussionWhat does a subnet participant actually control?
Toufic examined control of keys, operations, emissions and decisions, according to the organizer’s recap. His session asked what ownership means in practice for subnet participants. It did not supply a universal legal ruling on every subnet or validator arrangement.
The Shizzy takeA token balance, a signing key and an interest in a company can represent different rights. Clear explanations from teams help people understand what they are joining and who makes consequential decisions.
SN60 Bitsec
Security presentationAudit the reward system itself
John Yu discussed SN118 Ditto incentive findings and seed timing. The organizer attributes a fix statement to him. SN60 Bitsec separately announced Sentios’ continuous incentive-mechanism audits, extending its offering beyond smart contracts. That fix statement does not resolve every separate SN118 Ditto operational issue.
The Shizzy takeThe contract can work exactly as written and still reward the wrong behavior. Testing how competitors can game an evaluation is valuable work for a network built around incentives.
SN64 Chutes + closing panel
Builder perspectiveProtocol progress has a development cost
In the closing discussion, Jon Durbin described the work behind SN64 Chutes’ adoption of conviction lock. He called the result safer while explaining how protocol changes can interrupt product development. The panel also brought together Metanova, TAO.com and Latent Holdings, moderated by Yuma’s Greg Schvey.
The Shizzy takeShipping an upgrade is only part of the job. Builders must absorb it into real systems. That perspective matters when judging how fast the ecosystem should change and what support teams need.
Yuma + Crucible Labs + Bitstarter
Founder-support discussionBack founders who can adapt
Representatives of Yuma, Crucible Labs and Bitstarter discussed founder selection and support. The organizer emphasizes adaptability beyond the original idea. This was a discussion of how teams are assessed, rather than an announcement of new investments in each speaker’s portfolio.
The Shizzy takeThe first pitch rarely survives every customer conversation. A team that can listen, change direction and keep delivering may matter more than how polished its original deck looks.
SN54 Yanez
Product demonstrationAn agent needs permission before spending
Jose Caldera demonstrated Gate and Pulse: a $100 request was blocked by a $25 allowance, then authorized after the limit changed. The team’s product announcement describes phone approvals through YID. The demonstration establishes the presented workflow; it is not an independent security audit.
The Shizzy takeThis is a control people can understand immediately. Give the agent room to work, but require your approval when it crosses a spending boundary. Clear permissions will matter as agents do more consequential jobs.
SN87 Provenonce
Testnet milestoneKeep evidence of what an agent did
Will O’Brien presented the first testnet cycle and Assurance Observatory. SN87 Provenonce describes evidence from agent executions becoming institutional memory. The organizer specifies one validator and two miners, all operated by the team with test TAO. This remains a testnet milestone.
The Shizzy takeIf an agent makes a consequential decision, a team needs more than its final answer. Keeping inspectable evidence can help explain what happened. Independent participation and production use are the next steps to watch.
Yuma · ASMR
Research proposalDoes more liquidity always help?
Youssef El Moujahid presented Alpha Subnet Market Reloaded, or ASMR. The published clip describes liquidity adjustments using slippage benchmarks and a volatility limit. Yuma announced a preview followed by fuller analysis in October. The presentation did not activate a new market mechanism.
The Shizzy takeLiquidity shapes how a trade affects a market. A proposal that tests those effects is worth reading carefully, especially its assumptions and trade-offs. We need the analysis before judging the outcome.
SN80 OpenRoboto
Product and business-model demoReal work becomes robotics training data
Cameron Wang showed the device behind Shift. His explanation connects data sales to robotics-model development, with a buyer’s temporary exclusivity followed by the data returning to the open training pool. This adds business-model detail to the existing product announcement.
The Shizzy takeRobots need examples of useful work. Connecting data collection to buyers could help pay for more of those examples. The commercial terms and actual customer demand will decide how well that loop works.
SN102 ConnitoAI
Requests open · marketplace waitlistTurn training requests into reusable experts
Isabella presented reusable experts built from training requests. The organizer reports requests open for review and a model-marketplace waitlist. That gives a clearer picture of the proposed Training as a Service workflow; it does not establish a generally available marketplace.
The Shizzy takeCustomers care about solving their own task. A training request is a concrete starting point. Reusing useful experts could make the work more valuable, provided quality and access are clear.
SN10 Pareton + SN53 engy
Partnership announcedOptimization meets an inference provider
Xavier Lyu announced the collaboration to improve SN53 engy’s inference latency and throughput with SN10 Pareton’s optimization work. The organizer published the announcement. It sets a clear technical goal; new joint production measurements and commercial terms were not established in the reviewed sources.
The Shizzy takeThis is the kind of connection worth following: one subnet’s work helping another deliver a better service. Measured improvements for actual users will tell us whether the partnership delivers.
SN81 Reliquary
Early-access invitationPost-training around customer tasks
The SN81 Reliquary team invited subnet teams into customer-task post-training, according to the organizer. Const’s subsequent explanation describes miners finding useful model rollouts that become training inputs. The invitation is an early-access step, rather than proof of a newly signed paying customer.
The Shizzy takeA generally capable model still needs to perform the job a customer actually has. Bringing real tasks into post-training could help close that gap. Better results on those tasks are the evidence to look for.
SN105 Beam
Studio live · capabilities evolvingStudio and tools for teams to build with
Nic introduced Studio, developer tooling and an Action Registry for extensions. SN105 Beam directly confirmed Studio’s availability and published integration guidance. Its reviewed documentation distinguishes supported transfer and Room features from tunnels and a browser SDK that were still pending.
The Shizzy takeA network becomes easier to use when developers get a clear path from an idea to a working integration. Good tools and honest capability documentation matter as much as the demo.
SN85 Vidaio
Event deployment · product updateThe conference itself becomes a product demonstration
Gareth described the multilingual conference stream. SN85 Vidaio’s own recap confirms beta-testing its streaming product across the event and launching Item Removal that week. The summit supplied a real setting for the team to show its video tools.
The Shizzy takeThe audience could use part of the product while hearing about it. That is a strong way to make an infrastructure pitch concrete. Reliable performance and repeat customers are the steps that turn a good event demo into a business.
SN118 Ditto
Product presentation · roadmapShared context, portable memory and a Router track
Peyton Spencer and Omar Zanji presented V2’s context and memory workflow. SN118 Ditto’s own recap also outlines bespoke business deployments and a planned Router mining track for cost-efficient coding workflows. The team promised further details; the presentation does not establish that every proposed track is operational.
The Shizzy takeThe useful promise is continuity: less rebuilding context each time someone changes tools or starts another task. Pair that with quality-preserving cost reductions, and there is a clear customer problem to solve.
SN110 Green Compute
Planned expansionAnother data centre in the pipeline
Josh Riddett announced another $4 million data centre targeted for the end of October, after a $3 million facility he says recently went online. These are presenter-reported figures and timing. The new facility was described as a future deployment.
The Shizzy takeCompute needs physical machines, power and customers. A quantified build plan gives us a milestone to track. The next update should show when capacity comes online and what demand it serves.