After the number of developers was cut: Cripto didn't die, just gave it to AI
Author:Xinyang & EthanIOSG
In 2026, the GitHub dynamic curve of the Crypto open-source community completed an amazing "delay". From the peak of 2022, 45K monthly live developers returned to about 23K, halving this paper-based data and triggering social media discussions on “dissemination”. However, when we break down the cross-section of this curve, it is not the contraction of the industry, but a deep “brain to leverage”。
▲ Data source: Ericto Ecosystems Github
- Who's gone? Who's left
It's mostly new people who are leaving. The number of new developers in February alone reached 5462 in February 2024, followed by a significant decline, resulting in a loss of less than one year. Most of these people are in the cattle market, doing the NFT casting contract, the Fork DeFi protocol, doing the front end for the new L2. These jobs are highly dependent on the heat of the market, as soon as the heat passes, the project ceases to operate and the jobs disappear. In terms of data, new people never contributed more than 25 per cent of the total, and they were not at the core of the industry from the start。
▲ Newcomers marched with cattle and bears; Established devs (2 more years of experience) reached record high in the same period
Source: Electric Capital Development Report
On the other side, developers who have been in the line for more than two years have not gone up in the same period, making a record high, contributing about 70% of the code volume. The judgement of the GP Maria Shen of Electric Capital is straightforward: "When we look at this group, it grows and looks very healthy."
They did not stay because there was no other choice。
technically, the core work of the crypto is now infrastructure development that generally takes years to understand: protocol layer development, security audits, cross-chain architecture, which takes years to build up to be truly effective, and that cannot be eliminated from the market without the heat。
Economically, a lot of veterans have token and protocol power and equity relationships, and their accumulation in this industry has created real barriers and rewards. In terms of ecological distribution, they vote by foot: Bitcoin developers grow by 64.3 per cent in two years, Solana + 11.1 per cent, andCosmosDecline 51.1 per cent and Polkadot 46.9 per cent. Veterans are concentrating on eco-centres with real users and income away from projects that are still being maintained by narratives。
Source: Coincub Web3 Jobs Report 2025
Data source: Web3. Career
The same thing is confirmed by changes in the structure of posts. Of the new Web3 posts in 2025, the highest share was not for developers, but for Project & Programme Management, more than 27 per cent. This is counter-intuitive for a technology-driven industry, but the logic behind it is not complicated: the industry moves from the construction period to the implementation period, more than 100 chains need to be integrated, institutional clients come in and require completely different compliance and security requirements, and DAO governance requires a balance between different stakeholdrs. This is not a traditional project management exercise, but rather coordination and judgement in an environment where rules are still being developed。
The industry surface is shrinking and core density is increasing. In 2018-2019, the bear city was also accompanied by a massive loss of developers, but it was followed by Uniswap, AaveOpenSeeThe phenomenon level projects define cattle markets in 2020-2021. This round of builders has more mature infrastructure, and the AI era has given them a bigger stage than the last。
II. What are the abilities of those left behind
Crypto, what are the special abilities that the industry has developed in the builder? To answer that question, we need to go back to the bottom principle of the block chain, between the rotational cycles of the bear and the bear, a business that always operates on the same bottom rule: code, law, enforcement, final。
In 2016, the DAO incident, the attackers used a back-to-back loophole to transfer $36 million. The code does not exist, the logic is fully implemented as expected, except that the boundary was not anticipated by the designer. 2021PolyNetworkcross-chain bridge was attacked and $610 million was transferred within hours. there is no platform to stop, no institution to withdraw and no legal provisions to recover. this is the structural characteristic of crypto, which is distinguished from almost all other industries: faulty space is zero, and aftercare interventions are almost non-existent。
This environment is driven by a set of capacities that are rarely needed in other industries:In the absence of rules and trust, a functioning system is built from zero to allow strangers to participate。
This capability has two dimensions. One is to build trust from zero, not relying on any external authority, but relying on codes and mechanisms that allow strangers to be willing to put real assets in. The second is to judge under the dual uncertainties of technology and economy, without a regulatory framework, with no historical data, with no industry standards to refer to, and still with functional systems。
both levels have specific validations in crypto。UniswapWITHOUT CORPORATE GUARANTEES, WITHOUT KYC, WITHOUT CUSTOMER SERVICE, ANYONE WHO PUTS MONEY INTO A LIQUIDITY POOL DEPENDS ON TRUST IN HUNDREDS OF LINES OF CODE AND AN ECONOMIC MECHANISM, MAKING BILLIONS OF DOLLARS A DAY。MakerDaoDAI ' s stability is maintained by purely chain governance and collateral mechanisms without central bank endorsement and deposit insurance. DeFi Summer was even more extreme, with no regulatory framework, no auditing standards, no historical data to refer to, and the builder designed AMM, loan agreements, mobile mining, from concept to billions of dollars in TVL, for only a few months. This ability is expressed in different forms on the protocol, application and governance layers, but the underlying principles are the same。
AI is creating a highly structurally similar problem. Model decision-making is not transparent and output results cannot be independently validated. AI angent began to carry out transactions on its own initiative, allocate funds, and the accompanying system of rules and binding mechanisms did not yet exist. Large model companies control both models and assessment criteria, and users lack effective certification tools. The calculus is highly concentrated in a small number of top-level large plants, with monopolistic pricing occurring when demand erupts. These questions point to the same core: the question of trust in the autonomous system, which is repeated in the larger AI process。
crypto Builder has been dealing with such issues for many years in an environment where there are no rules of external authority, except that the scene was a chain agreement and now it is AI. And a group of people have brought the cumulative capacity of the crypto directly into AI and have run out of results。
III. HOW ARE THESE CAPACITIES RE-PRICING IN THE AI ERA
The cases of transition from crypto to AI have been frequent in recent years, but they have taken different things apart。
One of the most visible routes is the direct smoothing of hardware and experience. The three founders of CoreWeave, Michael Intrator, Brian Venturo and Brannin McBee, have been digging Ether from GPU since 2017, extending from one to thousands of mines in 2022, closing mining operations in 2022, publishing ChatGPT two months later, the GPU in hand became a direct AI calculator, coming to market in NASDAQ in March 2025, with the IPO estimated at approximately $23 billion, and the market value has since peaked at almost $70 billion。
The co-founder of OpenSea, Alex Atallah, has dealt with the problems of aggregation and route of highly exotic assets in the NFT market, moving the same set of experiences to the AI Model Route, creating OpenRouter, serving more than 5 million developers in two years, valued at $500 million。
Another type of migration is even more noteworthy. The NEAR founder, Illia Polosukhin, was one of the co-authors of the Transformer paper, who, after leaving Google that year, initially wanted to construct AI applications in natural languages, but faced a real problem in the development process: the need to mark cross-border payments to workers around the globe, most of whom do not have bank accounts, and block chain technology is the best solution to this payment problem。
The NEAR is now being transformed into an AI infrastructure platform with the core orientation being user-downed AI and decentralised confidential machine learning (DCML), allowing users to use AI services without exposing data. The experience of decentrization in the NEAR has become the most difficult starting point in this direction。
Circle co-founder Sean Neville, who left, created Catena Labs to locate the AI Native Bank, to migrate the understanding of the stablecoin infrastructure directly to the AI environment, a16z crypto led 18 million seed ships. The senior developers of Aave and Lens Protocol, Nader Dabit, turned to Cognition, bringing the developers ' ecological construction experience accumulated in a number of crypto protocols into the area of the AI instrument。
THE GROUP TOOK AWAY NOT JUST GPU HARDWARE OR USER NETWORKS, BUT THE INTUITIVENESS OF THE DESIGN OF THE MECHANISM, THE DEVELOPMENT EXPERIENCE OF THE DEVELOPER ' S ECOLOGY, AND THE JUDGEMENT TO BUILD A CREDIBLE SYSTEM FROM ZERO IN THE ABSENCE OF RULES. THESE CAPABILITIES CORRESPOND TO THREE STRUCTURAL GAPS IN THE SIZE OF AI。
Calculus aggregation and optimization
Calculus is the most direct bottleneck of AI scale. Training and reasoning require large quantities of GPUs, volatile demand, expensive cloudworks and queues, and businesses do not want to hoard hardware themselves. There are two dimensions to the problem: how computing aggregates the distribution and how aggregates can work more efficiently. Crypto Builder has a direct transferable accumulation at both levels。
Hyperbolic solves problems of distribution and trust. The founder, Jasper Zhang, brought the decentrized mechanism into the AI Calculus: Token, allowing dispersed GPU holders to contribute to idle computing, but the more central problem is trust。
How can one believe that a strange node gave the correct result? The core innovation, the PoSP, uses random sampling gab theory to make honesty the preferred strategy of nodes without full validation, low cost, scalable and reliable results. This mechanism moves directly from the logic of crypto to verify the behaviour of a strange node。
MoonMath solves problems of efficiency. The predecessor, Ingonyama, focused on the acceleration of ZK hardware, increased the ZK certificate generation rate several times under extreme computational constraints. The direction is now directed to the Physical AI performance layer, where the thin focus acceleration (LiteAttention), the low decomposition (LiteLinear) of the FNN layers, and the training reverse transmission acceleration (BackLite). From ZK to AI speeds up the reasoning, the bottom is the same set of abilities: to make math run faster under extreme computational constraints. The track has changed, accumulation has not been wasted。
AI GOVERNANCE AND INCENTIVE DESIGN
How to ensure that multiple AI parties start working together to implement their mandates and that they do not undermine the whole system while pursuing their respective goals. Each participant is pursuing its own target function, and no one is sure that the system will function when they are added together, and angent is operating at a much faster rate than the window of human intervention。
This is the type of problem that the crypto builder has repeatedly addressed in the DAO governance and design of tokenomics: to allow participants with completely different interests to operate without central authority in the intended direction of the system. The answer given by crypto is an economic mechanism in which irregularity has a real economic cost, and the rules are written in code and automatically enforced。
EigenLayer moved the mechanism directly to the AI scene. Through the restaking mechanism, node needs to pledge assets before engaging in collaboration, and failure to comply or breach would trigger automatic punishment, the rule is not a recommendation, but a rigid border with real economic costs. EigenClaud extends this logic to the validateable computing and collaborative governance of AI anent, so that angent has to fall within the pre-set when pursuing his own goals. Economic mechanisms to bind an individual are much more reliable than ethical ones。
AI Agent Automatic Payment
there's a more fundamental question: how angent pays. traditional payment systems are designed for people, credit cards need to be opened, bank transfers need to be authorized, and each step assumes that the operator is human, has an identity and is waiting. angent will not wait, and it may launch a large number of requests per second, each of which may involve micropayments, and the traditional payment pipeline will directly lapse in this scenario。
Stablecoin and the chain rules are the infrastructure that crypto builder has already built, and the original supports programmable, non-authorized, 24/7. These three features happen to be the mandatory requirements of angent paying for the scene, and all that's missing is a layer of stablecoin that gets an agreement in the angent workflow。
x402 Launched by Coinbase in May 2025, activates the HTTP 402 status code, sets the servicecoin payment directly embedded in the HTTP request, angent initiates the request and completes the payment, without account, for about two seconds. As of April 2026, the X402 agreement had processed more than 165 million transactions, with a cumulative volume of approximately $50 million, and 69,000 live parties (data source: x 402 Foundation), Cloudflare, AWS, Stripe, Anthropic MCP were connected. Angent's paying has been a track with real traffic。
THE THREE DIRECTIONS CORRESPOND TO THE THREE STRUCTURAL GAPS ENCOUNTERED IN THE SIZE OF AI:compassion and efficiency, incentives for multi-agent collaboration, and autonomous payment infrastructure。these three questions have no available answers in the traditional software architecture, but there are corresponding treatment experiences in the crypto industry. the power didn't disappear, just found a new carrying scene。
IV. The new location of Builder: from the person who wrote the contract to the person who set the rules for AI
The scaling up of AI is creating a functional gap that previously did not exist. It is not a shortage of skilled personnel, but a gap for those who are able to design confidence mechanisms in an autonomous system. The role of the client is being redefined when the object of the service changes from person to AI, crypto builder。
The following table compares the dimensions of the specific functional paradigm:
The core difference between the two paradigms lies not in the technology grid, but in the way trust is built and the logic of implementing rules. Pre-AI age, crypto builder faces human participants, with rules written into contracts, with no wrong space, but with relatively clear boundaries。
In the AI-Native era, when the interactive object becomes an AI agent operating autonomously, the problem that needs to be addressed is: the behaviour of angent is unpredictable, execution is far faster than the human intervention window, and the boundaries of the system itself need to be redefined with greater uncertainty. The functional positioning of crypto builder is moving from “writing safe contracts” to “designing credible mechanisms for AI autonomous systems”。
The recruitment of front-line agencies already reflects this change:
2026 Q1 HEAD EXCHANGE ACTIVE AI/DATA CORE POST
Source: Gate Research Institute
Recruitment by front exchanges and agencies in 2026 clearly reflects this trend:Instead of simply recruiting AI engineers or crypto developers, looking for people who can connect the two sides, who both know the chain of incentives to distort and govern the game, who can embed the AI tool deep into the crypto workflow, and who can design mechanisms that allow angent to align with regulators and users over time。
The direction of capital allocation has also reflected this judgement. Paradigm is raising a new fund of up to $1.5 billion, extending from crypto to AI and robotics. Haun Ventures completed $1 billion Fund II, focusing on the financial infrastructure integrated with AI, in particular the payment, stabilization and economic system that supports the autonomous trading and coordination of AI entities。
a16z crypto completes the $2.2 billion Crypto Fund V, making it clear that the Fund will invest 100 per cent in the crypto area. Faced with the complexity and opacity of the AI era, they will focus on the direction of the application of the crypto ' s transparency, certifiability and decentrization. Also, according to PitchBook, about 40 percent of the U.S. Crypto VC investments in 2025 went to AI-related companies, up significantly from 2024
Similarly, crypto bilder turned to AI, and the path chosen in different market settings varied markedly。
In the United States, with relatively clear regulatory environments, agreement-level innovations have gained real space for survival. Capital network density is high, with short paths from ideas to financing and greater room for error. The common feature of projects such as Hyperbolic, EigenClaud, Gensyn and Ritual is the design of new mechanisms from zero rather than simple application integration on existing systems. Top VC has clear investment papers in the direction of “valitable computing, Agent coordination, decentrization ML” and is willing to provide sufficient tolerance for early technological exploration。
The situation is different in Asia. Singapore and Hong Kong have assumed more of the role of compliance and institutional finance, with relatively conservative regulatory frameworks and less tolerance for purely conventional innovations. More applications and industry integration paths - fast access to AI products and services using the user base, ability to pay or data assets accumulated by crypto - are selected when the builder turns AI with the crypto background。
This is not a capacity gap, but rather a difference in path choices caused by different market signals and regulatory environments:The United States encourages bottom-up innovation and early technological exploration, while Asia places greater emphasis on compliance-friendly, fast-realization and deep linkages with traditional industries。
Go back to the first GitHub curve. The monthly developer dropped from 45K to 23K, ostensibly an industry shrinking. But of the people left behind, stablished dev is at an all-time high, is rushing to the ecology of real users, and is being re-priced in an unprecedented way by the AI industry。
When AI scale encounters structural bottlenecks, such as calculus aggregation, Agent self-payment, data and decision validation, privacy coordination, these Builders, at the node where Crypto meets AI, have accumulated a sense of sensitivity to rules, incentives and authenticity, and are gradually turning into a system-level capacity that is scarce in the AI era。
As an investment agency in the crypto infrastructure starting in 2017, IOSG's judgement on this line goes beyond observation. We have been involved in the EigenLayer restaking mechanism before it was widely recognized in the market, taking over the Ingonyama (now MoonMath) seed-wheeled ZK hardware to accelerate the migration to the AI performance layer, and investing in Hyperbolic in 2024 to see the path of using the crypto original certification mechanism to solve the problem of de-centrization algorithm trust。
The common logic behind these layouts is that the trust, coordination and validation problems encountered by AI in scaling up will ultimately require institutional design capabilities accumulated by the crypto industry。We believe that the intersection of crypto and AI is not narrative, but a structural opportunity that is taking place。
