4 Years of Beanstalk
On the fourth anniversary of the initial Beanstalk deployment, the potential for a credit based money to free crypto from the existential threat of centralized stablecoins has never been higher. If you haven’t been following along, and want to get up to speed on the most promising experiment at the frontier of money, this piece is for you.
Before Beanstalk - Empty Set Dollar
Prior to the deployment of Beanstalk, there had been a variety of attempts at creating a credit based algorithmic stablecoin, starting with Basis in 2017. The most successful model through 2021, Empty Set Dollar (ESD), suffered from a number of problems.
Instead of letting the market naturally discover the right interest rate, ESD imposed an arbitrary one on participants. The debt issuance mechanism used an interest rate curve that (1) set the interest rate as a function of the debt level, independent of what the market was willing to pay, and (2) had a hard cap, which led to major inefficiencies in the protocol’s ability to borrow excess supply from the market: everyone wanted to wait until the maximum interest rate, but because the interest rate was a function of the debt level, the protocol would go through extended periods of time with a stalled interest rate and no lending.
The fungible nature of ESD's debt created perverse incentives which dissuaded people from being the first to lend to it; you could lend to the protocol last and redeem first, if you were the fastest to redeem. This structure, combined with the fact that as others lent to the protocol the interest rate would increase, made the optimal strategy to wait until after others had already lent to the protocol and favored whales with better execution.
ESD had no bank run prevention mechanism, which led to excessive downward price swings.
While ESD was a pioneer in its own right, its inefficiencies led to its, and its many forks', quick demise. All ESD forks completely collapsed within ~2 debt cycles and a maximum of a few months.
Beanstalk - A New Model
Beanstalk was inspired by ESD, and implemented novel incentive designs that remedied the three main inefficiencies of ESD.
The protocol introduced a first of its kind on-chain PID controller that automatically adjusted the interest rate on loans without a predetermined interest rate curve, allowing for natural price discovery. Although slightly oversimplified, in general if there was not enough demand for debt, Beanstalk would raise the interest rate; if there was excess demand for debt, Beanstalk would decrease the interest rate.
The first in, first out debt repayment mechanism, Pods, leveraged the trustless nature of smart contracts to create a novel type of debt that created a radically improved lending market that fostered the protocol successfully borrowing from the market during periods of excess supply. Because those who lent to Beanstalk first were the first to be paid back when the protocol grew, once someone felt like the interest rate was sufficiently high, they were incentivized to lend to the protocol or risk someone else getting in line before them.
The Stalk System created the first ever permissionless bank run minimization mechanism that leveraged a time-weighted incentive scheme to align long-term holders with the protocol while maximizing the distribution of ownership in the protocol. Whereas in previous systems, the efficient behavior during a bank run was to leave and come back later, independent of how much one believed in the protocol, the Stalk System rewarded holders for leaving their Deposits in Beanstalk during tough times.
Beanstalk Live - The First 8 Months
Although Beanstalk introduced major improvements over its predecessors, there were a variety of inefficiencies in the initial version of the model that made the protocol unsustainable. The first 8 months were spent plugging holes on a sinking ship.
Beanstalk was willing to issue excessive amounts of debt – more than was necessary or efficient – such that the debt level of the protocol skyrocketed. The interest rate the protocol was willing to pay increased even when it wasn’t optimal to do so, and the way the protocol measured demand for debt was flawed, both of which led to Beanstalk overpaying for loans. Issuing too much debt and paying too high an interest rate were both unsustainable.
Furthermore, there was no way for Depositors in the system to use their Deposits to contribute to protocol maintenance, despite being the participants and value in the protocol best suited to do so. Until the introduction of Converts, large amounts of capital sat stagnant within the system, contributing to excess volatility.
While Beanstalk was far from perfect, it was good enough that there was enough time to observe its inefficiencies and fix them. The protocol bent, but it did not break.
Beanstalk Exploit - A Shakespearean Tragedy
In its ninth month, something incredible happened: the protocol actually started to work. Enough improvements were made, and enough holes were plugged, that the ship started to float.
The protocol entered its healthiest and most promising period, paying off its debt and increasing its liquidity. The Beanstalk community Discord was electric, and Twitter started to seriously pay attention to what was occurring. For about 3 weeks, it looked like credit based money was about to shake up the cryptocurrency landscape.
But shortly after the protocol started to run, tragedy struck: a governance exploit, that ironically was only made possible due to the excessive health (i.e., liquidity) of the protocol, was executed by a still unknown attacker, and drained all ~$77m of liquidity from the protocol, destroying an additional ~$120m in protocol-native value in an instant.
Because the protocol had demonstrated efficacy in the weeks immediately prior to the exploit, the community organized to revive and restart the protocol. However, 3 weeks after the Beanstalk exploit, Terra Luna, the largest algorithmic stablecoin up to that point (and still today) was destroyed via an economic and social attack that destroyed $60B of value in less than a week, effectively ending the crypto bull run and eviscerating almost all desire amongst the crypto community to continue to pursue algorithmic stablecoins.
Beanstalk Replant - Trauma Unresolved
Beanstalk was replanted on August 6th, 2023, a year after its initial deployment. While the protocol was able to borrow ~$17m in new value from the market as part of its recapitalization, the structure of the recapitalization honored all outstanding Beanstalk liabilities at par.
In an article shortly before the Barn Raise to recapitalize the protocol, Publius wrote “One of the main attractions of the more aggressive Barn Raise strategy laid out in BFP-72 **is that Beanstalk is likely being Replanted with a ridiculously high Pod Rate in the worst a) macro environment in at least a decade, b) crypto market in years, c) stablecoin market of all time, and d) endogenous circumstances possible, as a result of the attack. This presents an incredible opportunity for the model to demonstrate its efficacy. We will all know very quickly if it is working or not.”
Although the protocol did “work” after it was replanted, continuing to cross the value target for another 10 months, between the negative publicity associated with the exploit, the general fear of algorithmic stablecoin after the collapse of Terra Luna, and the system's excessive debt, interest all but disappeared.
A Data Deficiency
During the first nine months of Beanstalk’s life prior to the exploit, there was so much data collected that there was a significant number of improvements to the mechanism that were imagined – so much so that there was enough development work to continue iterating on the model for another 2 years after the Replant.
However, without users actively participating in the system, our ability to (1) learn about the efficacy of the improvements that were being implemented and (2) discover new ways to improve the model, was compromised.
While the model had come a long way from the early Beanstalk days when the ship had so many holes that it barely survived, there was a lot of reasons to believe that the model was far from an optimal state. It just wasn’t clear what were the next steps to take the model to the next level.
Pinto - Debt Restructuring and Data Dump
Enter Pinto. Pinto was deployed in November of 2024 with a variety of model improvements – the last set that was reasonably conceivable given the data collected through Beanstalk. Pinto restructured the Beanstalk debt to honor it in the instance where the protocol succeeds (Pinto will start paying 3% of all mints to Beanstalk after it reaches a 1 billion supply) but effectively gave the model a clean slate to collect more data.
In the 8 months since it was deployed, the protocol has experienced a period of rapid growth followed by its first debt cycle, extended time below the value target, and a return to it. This broad swath of data has enabled the protocol to be improved further and is enough to enable at least another year’s worth of development. Expect a roadmap in the near future.
4 Years of Beanstalk - Perspective
Despite over 35,000 sunrises in the two protocols, it hasn’t been all sunshine and roses. Things certainly didn’t play out how we drew it up on the big board almost 5 years ago when we started working on the project; it’s hard not to fall into thinking about “what if” Beanstalk had never been exploited. Maybe all the money that fled Terra would have flown into Beanstalk, and the protocol would have grown into the billions in short order.
But, the glass half full says that the Pinto model today is in a dramatically better spot than Beanstalk was at the time of the exploit, and it’s only a matter of time before the crypto market reconsiders the potential for algorithmic stablecoins.
Building a network-native algorithmic stablecoin that can sustain itself perpetually may be impossible. Without one, however, the security of crypto networks will be permanently compromised. We must find out how to create one, or die trying. The model is ready for its next major test.
See Ben's original post from August 6th, 2025 here.
Last updated