# Into: Byzantium Byzantium turns marketing into a mineable commodity: brands fund a campaign, AI agents compete to drive real clicks, and the network pays out only for the clicks that survive a fraud check. // Marketing as a mining problem --- > New to Bittensor? Start here. Experienced users can skip to the full analysis. ### What is Byzantium? Byzantium (Subnet 76) is a decentralized AI marketing agency built on Bittensor. Brands post a campaign with a budget, goals, and guardrails; independent operators run AI agents that promote the brand across social platforms; and the network pays based on the genuine clicks those agents drive. According to the project's website, it launches on X and Farcaster, with more channels planned. **The simple version:** It's like hiring a performance marketing agency, except the agency is an open market of competing AI agents, and you only pay for clicks that pass a fraud check. **Centralized equivalent:** A performance marketing or influencer agency, or a social ad network. The difference is that no single agency takes a retainer: anyone can supply agents, and budgets release against verified results. **How it works:** - **Miners** run autonomous AI agents that craft posts, replies, and threads across social platforms, competing to drive genuine clicks for a brand's campaign. - **Validators** measure the genuine clicks each miner produced, filter out bots and spam, and set the weights that decide rewards. Per the site, scoring is meant to be independent and recomputable, cross-checked against public platform data. --- ### Why This Matters - **The problem it solves:** Digital marketing spend is plagued by bots, fake engagement, and opaque agency retainers. Byzantium's pitch is that you fund a campaign and the protocol pays only for clicks that survive validator scrutiny, so spend tracks real attention instead of vanity metrics. - **The opportunity:** Performance marketing is a large, globally distributed activity that maps cleanly onto a competitive market of agents. If verification holds up, marketing becomes another kind of work that can be mined. - **The Bittensor advantage:** Marketing is a measurable, adversarial task. There is a real output (clicks) and a real failure mode (fake clicks), which is a natural fit for an incentive network where miners compete and validators police the result, rather than a single agency grading its own homework. - **Traction signals:** Early. The subnet's own site shows zero agents deployed, zero clicks driven, and zero active miners as of writing, and it is running an early miner access waitlist. No public usage data exists yet, and a subnet-specific search turned up no independent coverage. This is a pre-launch network, not a proven one. --- ## Full Analysis **Category:** Other (AI Marketing and Attention) | **Centralized Competitor:** Performance marketing agencies, influencer platforms, social ad networks Most attention you can buy online comes with a trust problem: bot farms, fake engagement, and agencies whose incentives do not fully line up with the brand paying them. Byzantium's framing, drawn from its website, is to replace the agency retainer with an open market where the protocol pays only for verified results. The name nods to the Byzantine generals problem, the trust-coordination puzzle at the core of every decentralized network. **Mechanism:** The flow described on the Byzantium site has four steps. A brand posts a campaign with goals, target audiences, and guardrails, and escrows a budget that releases against performance. Miners run autonomous AI agents that produce posts, replies, and threads across social platforms, competing to drive genuine clicks. Validators then measure those clicks, filter out bots and fakes, and score on quality of signal rather than raw volume. The subnet distributes its alpha token to the best-performing miners, proportional to the genuine clicks their agents drove. Supported platforms are X and Farcaster at launch, per the site's FAQ. A note on how rewards actually flow, since it is a common point of confusion on Bittensor: validators do not pay miners directly out of campaign budgets. They set weights, and the protocol's emissions follow those weights. Both TAO and the subnet's alpha token are injected into the subnet's liquidity pool each block, and at the end of each tempo the accumulated alpha is distributed to participants under a fixed protocol split: 41% to miners, 41% to validators and their stakers, and 18% to the subnet owner. The campaign-budget product sits on top of that incentive base. Two caveats are worth stating plainly. First, the hardest part of this design is verification: telling a genuine click from a sophisticated fake is an adversarial problem, and the site describes the approach (recomputable scoring, cross-checking public platform data) without yet publishing the detailed methodology. Second, Byzantium has not listed a public code repository. The website marks both its docs and GitHub as coming soon, and no repository is registered to the subnet's on-chain identity. That means the mechanics above are sourced to the project's own website, not to inspectable code, and the specifics could change as the network goes live. On-chain, the picture is consistent with a very early subnet. As of June 14, 2026, the alpha token trades around 0.00354 TAO with roughly 1,329 TAO of depth in its liquidity pool, and the subnet's current emission share is 0%. Under Bittensor's flow-based emission model, a 0% share reflects net staking outflows rather than inflows, which for a pre-launch subnet with no live product is not surprising. --- ### Risk Factors - **Deregistration risk as immunity ends:** Byzantium registered on February 19, 2026, so its 4-month network immunity period lapses around June 19, 2026. After that, a subnet sitting at a 0% emission share and a low EMA price becomes a candidate for automatic deregistration when a new subnet registers. Reaching live usage and positive net flows before then matters. - **Verification is the whole game:** The model only works if validators can reliably tell genuine clicks from fakes. That is an adversarial, evolving problem, and the detailed scoring methodology is not yet public. - **No public code or docs yet:** No repository is registered on-chain and the site lists docs and GitHub as coming soon, so the design cannot yet be independently audited. The claims here rest on the project's own description. - **Unproven traction:** No agents, clicks, or brands are live yet by the project's own counters, and execution risk is high for a new team building a novel verification problem. --- Into the next one.