Researching what’s happening in DeFi, why it matters, and how to design better systems
New York, NY
- Tweets 147
- Following 132
- Followers 570
- Likes 533
Designing DeFi retweeted
🚨New Paper with @Weiye_Xi , @ciamac and @Qiaoqiao2001
Here’s two different prediction markets priced at ~6.5%:
1. Will the US confirm the existence of Aliens in 2026?
2. *that* Spurs @ Knicks Game 4, ~4th quarter.
They suggest that both these events are the same probability, but intuitively these feel very different: the latter (Knicks won!) feels a lot more uncertain. But prediction markets don’t immediately give us a way to quantify it.
7
17
99
32,279
At D², the AMMs track was about market design beyond swap execution
The talks covered how automated pricing rules and market workflows can shape:
→ Energy coordination
→ Prediction-market pricing & resolution
→ Perp open-interest balance
Read more: substack.com/home/post/p-204…
2
1
23
4,092
Hyperbolic Funding Rate Mechanisms
Presented by: Nihar Shah
Perp markets need funding rules that respond to capacity. This talk showed how nonlinear funding can stay mild near balance and push harder as open interest imbalance grows
1
1
188
SoK: Market Microstructure for Decentralized Prediction Markets
Presented by: @PulpSpy
Prediction-market prices depend on the full workflow: market creation, trading, resolution, settlement, and archiving. The process shapes what the price actually means
1
2
7
232
Under stress, derivatives protocols have to make hard decisions in real time:
→ How options decay
→ How leverage is reduced
→ How ADL is triggered
→ How losses are allocated
→ How solvency is restored
Full recap of the Perpetual Futures & Derivatives track at D²: substack.com/home/post/p-203…
3
4
19
1,677
Risk-Based Auto-Deleveraging
Presented by: Natascha Hey
ADL should reduce the risk that matters. The presentation showed how venues can target dangerous leverage first instead of relying on a simple queue
1
5
231
Autodeleveraging as Online Learning
Presented by: @0xnagu & @ConejoCapital
Stress events unfold in rounds. This talk looked at how ADL can adapt as prices, liquidations, and solvency needs change
1
1
11
639
Designing DeFi retweeted
📢 Call for Papers
We are organizing DePIN-Energy @ BRAINS 2026: a workshop on Markets for Energy Trading and Decentralized Physical Infrastructures.
📍 Florence, Italy
📅 October 13, 2026
📝 Submission deadline: July 15, 2026
🔗 Submit via EDAS: edas.info/N35433
2
7
932
As DeFi matures, more design questions sit in the microstructure layer:
→ Transaction ordering
→ Fee mechanisms
→ Gas metering
→ Block building
→ Geography
These infrastructure decisions shape execution quality, congestion, MEV, and credible decentralization
1
2
22
1,137
@The_Zhukeeper @DrissiFay @SarishtW @jason_of_cs & @syang2ng presented work in the Microstructure track at Designing DeFi 2026
Read more ↓
designingdefi.substack.com/p…
1
5
319
Designing DeFi retweeted
The second annual OMI-SBS Conference on FinTech & Blockchain Economics is happening June 25–26 at Oxford University
Join academics and industry leaders for conversations on the research and ideas shaping the future of financial markets
Learn more ↓
sites.google.com/view/omi-sb…
5
2
16
3,640
Jay designed DeFi
Had a ton of fun presenting at @Designing_DeFi my paper "Paper Agents, Paper Gains: An Empirical Study of DeFi Investment Agents", a survey tracking the scope, design, and performance of trading agents from Nov '24 to Nov '25 with @theamyzhao @sui414. Our key findings:
> We scraped 1900+ projects on CoinGecko (post-TGE) and Messari (pre-TGE) tagged as AI, filtered out pure infra/non-agent projects, and selected 10 representative agentic projects, incl. Giza, Surf, Numerai, AIXBT, Eliza.
> We evaluated their strategy, observability, source of funds, AUM. Most agents don't actually autonomously spot trade - either they provide investment advice or engage in passive yield strategies. Even if they do trade, they don't have great onchain observability and verifiability that an agent is actually signing (vs. a human signer).
> We chose to analyze Virtuals + ElizaOS frameworks deeper as these had the highest onchain standardization + observability. Looking at top volume projects built on Virtuals + ElizaOS, we realize that several have $100M+ peak market cap, but extremely high MC-to-AUM ratios, with some peak ratios exceeding 10,000x (compared to 1-3x for DATs). They mostly also trade micro-cap tokens, often holding other agents' tokens.
> We then choose 11 Solana projects to analyze onchain PNL for users and agent treasuries (rPNL + uPNL). There seems to a disconnect - treasuries have a net profit of +$34.3M, but users have losses of -$191.7M
> Furthermore, there's heavy user PNL inequality, 81.4% of gains are captured by top 1% users, while 62.2% of all users suffer net losses. Agent tokens had 93% peak to trough drawdown vs. Solana only had 54% drawdown. So for the average user buying into these DeFi Investment Agents, things didn't really turn out great.
> So where do we go from here? Our findings align with theoretical results abouts the "CoinAlg Bind", which posits a tradeoff between economic fairness and profitability for Collective Investment Algorithms (CoinAlgs). But DeFi trading agents are clearly here to stay, and the space is moving extremely quickly.
> In our view, a mature DeFi Investment Agent must satisfy 3 criteria - (1) verifiable autonomy/execution of trades, (2) sustainable profits vs. passive benchmarks, (3) stakeholder alignments, where profits are actually passed back to individual users. While the space is still early, we're really excited about how these DeFi Investment agents can continue to evolve.
[FULL PAPER LINK BELOW]
1
8
861
Designing DeFi retweeted
Enterprise-grade perps performance is here. What's still open is everything around it: how liquidity is provisioned, how funding is shaped, how losses are absorbed when things break, etc.
From @Designing_DeFi at Columbia by @Uniswap:
@NehharShah presented hyperbolic funding rate mechanisms as a solution for balancing open interest. Natascha Hey showed that the optimal ADL policy is to equalize leverage across positions and deleverage the most-levered first. @0xnagu and Mauricio Trujillo Ramirez frame ADL as online learning on a PnL-haircut domain and provided worst-case bounds.
ADL deserves rigorous mechanism design. The debate between prop AMMs vs orderbooks may even be settled by which are better at surviving big liquidation cascades.
4
4
24
966
