Cookie.fun

Summary

Cookie.fun was the flagship product of Cookie DAO, a project grown out of Cookie3, a "MarketingFi" analytics startup founded in 2021 in Tallinn by Filip Wielanier (CEO) and Wojciech Piechociński (CTO) and backed by roughly $5.5M from Animoca Brands, Spartan Group, and others. The COOKIE token (1B supply, deployed on BNB Chain and Base at 0xc0041ef357b183448b235a8ea73ce4e4ec8c265f) launched in June 2024. Cookie.fun itself went live on January 11, 2025, at the height of the AI-agent token mania, as an "index for AI agents": a terminal aggregating ~7TB of onchain and social data to rank agent tokens by "mindshare" — a normalized share of crypto-Twitter attention — alongside market cap, holder growth, and sentiment. In May 2025 it expanded into a full InfoFi platform (v1.0) with Cookie Snaps, a snap-to-earn creator marketplace that paid X posters for loyalty-weighted attention on partner projects, and in July 2025 added Attention Capital Markets (ACM) with Legion, tying token-sale allocations to earned attention. On January 15, 2026, X revoked API access for apps that financially reward posting; Cookie DAO shut down Snaps the next day and pivoted to "Cookie Pro," a market-intelligence product.

Design (Mechanism)

Cookie.fun layered several mechanisms on one data pipeline:

• Mindshare index. An AI pipeline scored each project's share of relevant crypto-Twitter conversation, weighting relevance, quality, engagement, and narrative alignment rather than raw post counts. Mindshare was zero-sum by construction: a project's percentage fell when total conversation volume grew, making it a relative attention market.
• Token-gated data access. Basic dashboards were free; premium analytics required locking 10,000 COOKIE on BNB Chain, Ethereum, or Base. Stakers also accrued Cookie Points redeemable for partner airdrops ("multi airdrop farming"), and tier levels (up to Diamond/Platinum) boosted campaign rewards.
• Cookie Snaps (May 2025). A relaunch of the earlier Cookie Affiliate KOL network (~20,000 creators, ~400M combined following in April 2024) as a project-to-creator campaign engine. Posts earned SNAPS via scores combining mindshare contribution, sentiment, consistency, and engagement depth, with a 10% lifetime referral share. Its differentiator versus Kaito's Yaps was a loyalty multiplier: rewards were weighted toward creators concentrating on few projects, and detected spam, copy-pasta, or coordinated boosting was slashed. Stakers received a separate 10–20% cut of campaign reward pools.
• Attention Capital Markets (July 2025). With Legion, cSnaps (earned attention) fed into merit-based allocations in onchain fundraises — "put your money where your mouth is," converting reputation into deal access rather than only emissions.
• Governance. Active COOKIE stakers formed Cookie DAO, which adopted a constitution in February 2025 as a step toward decentralizing control of the platform.

Outcome

Cookie.fun rode the AI-agent wave to become one of the two dominant InfoFi platforms (alongside Kaito). The May 2025 v1.0/Snaps launch onboarded over 56,000 users within a week on an open-access referral model, and the platform ran mindshare and capital-mindshare campaigns for dozens of projects through 2025. ACM launched with Legion in July 2025. The core dependency then failed: on January 15, 2026, X changed API policy to ban applications that financially reward posting, and on January 16, 2026, Cookie DAO announced the immediate shutdown of Snaps and all creator activities, stating "InfoFi is changing, and it is time to gradually sunset Snaps." The team pivoted to Cookie Pro, a real-time market-intelligence product slated for Q1 2026. The analytics/data business and token persist, but the signature attention-mining mechanism is dead — hence partial_success.

Why it worked

• Right product, right mania. Launching an AI-agent leaderboard in January 2025 gave traders a single legible metric (mindshare) for an otherwise unquantifiable narrative market; the terminal became default infrastructure for the sector.
• Loyalty-weighted scoring was a real mechanism improvement. By multiplying rewards for concentration and slashing detected farming, Snaps blunted the "shill every campaign" equilibrium visible in Kaito's Yaps, and its AI quality-scoring plus social-reputation layer made low-follower copy-pasta economically worthless.
• Stacked demand for the token. Data access, staking tiers, campaign reward cuts, airdrop points, and ACM boosts all pulled on the same COOKIE lock, aligning data consumers, creators, and speculators.

Where the design broke

• Total platform dependency. Every core loop — mindshare measurement, Snaps scoring, sybil detection — ran on X data via API. A single policy change by a centralized platform destroyed the flagship product overnight; no onchain fallback existed.
• Attention markets invite Goodharting. Despite loyalty weighting, observers described a farming environment where high point totals, not insight, were the goal; measured "mindshare" partly reflected reward-seeking rather than organic conviction, degrading the very signal the index sold.
• Zero-sum emissions treadmill. Paying for attention with token campaigns works only while new projects fund campaigns; when the AI-agent narrative cooled through 2025, the underlying demand for manufactured mindshare thinned.

Lessons

• Any mechanism whose measurement layer lives on a centralized platform (X, in this case) inherits that platform's policy risk in full; the January 2026 API ban ended an entire product category, not just one app.
• Loyalty multipliers and slashing are genuinely better anti-farming design than flat engagement rewards, but they mitigate rather than solve Goodhart's law — when attention is directly monetized, the metric stops being a clean signal of belief.
• Converting earned reputation into capital access (ACM's allocation model) is a more durable sink for attention scores than emissions, because it prices attention against real deal flow instead of inflating rewards.
• Bootstrapping via an existing creator network (Cookie Affiliate's 20k KOLs) can compress cold-start time for a two-sided attention marketplace dramatically.

Redesign (EDITORIAL — hypothesis, not fact)

This is editorial speculation. A rebuilt Cookie.fun would decouple measurement from any single social platform: score attention across an open-protocol substrate (Farcaster, Lens, onchain attestations) where reward-linked posting cannot be banned by fiat, and treat X data as one unverifiable input rather than the ledger of record. Creators would publish signed content hashes onchain so campaigns settle against censorship-resistant proofs of authorship. To fight Goodharting, replace direct pay-per-post with prediction-market-style skin in the game: creators stake on the projects they cover, and rewards scale with subsequent measured outcomes (holder retention, TVL, price-independent usage) rather than raw engagement — making mindshare a costly signal. ACM's attention-to-allocation pipe was the right endgame and should have been the core product from day one, with Snaps merely its reputation oracle.