Author:Xinyang & Ethan,IOSG
In 2026, the active developer curve on GitHub for the Crypto open-source community completed a remarkable "bottoming-out". Monthly active developers dropped from a peak of 45K in 2022 to about 23K, a halving on paper that sparked discussions about "narrative exhaustion" on social media. However, dissecting the cross-section of this curve reveals not an industry contraction, but a profound "talent deleveraging".
▲ Data Source: Electric Capital Developer Report, based on Crypto Ecosystems Github
I. Who Left? Who Stayed?
Mostly newcomers left. Monthly new developers peaked at 5,462 in February 2024, then dropped sharply, with a 52% churn rate for those in the industry for less than a year. These individuals mostly flooded in during the bull market, working on NFT minting contracts, forking DeFi protocols, and building frontends for new L2s. These roles were highly dependent on market hype; once the hype faded and projects ceased operations, the jobs disappeared. Data shows newcomers' code contributions never exceeded 25% of the total; they were never part of the industry's core layer from the start.
▲ Newcomers surged in with the bull market, left with the bear; Established devs (2+ years experience) hit a historical high in the same period.
Data Source: Electric Capital Developer Report
On the other hand, developers with over two years of experience didn't decrease but instead rose, hitting a historical high, contributing about 70% of the code volume. Maria Shen, GP at Electric Capital, puts it bluntly: "When we look at the established developers group, it's growing, and it looks very healthy."
They didn't stay because they lacked other options.
Technically, core crypto work now involves infrastructure development that typically requires years of accumulation to grasp: protocol-layer development, security audits, cross-chain architecture. This work takes years to truly master and cannot be washed away by fading hype.
Economically, many veterans hold unvested tokens, governance power in protocols, and equity relationships; their accumulated industry experience has formed real barriers and returns. Looking at ecosystem distribution, they are voting with their feet: Bitcoin developer count grew 64.3% over two years, Solana +11.1%, while Cosmos dropped 51.1%, and Polkadot dropped 46.9%. Veterans are concentrating towards ecosystems with real users and revenue, leaving behind projects sustained by narratives.
▲ Source: Coincub Web3 Jobs Report 2025
Data Source: Web3.Career
Changes in job structure corroborate the same trend. Among new Web3 positions in 2025, the highest proportion wasn't developers, but Project & Programme Management, exceeding 27%. For a technology-driven industry, this is counterintuitive, but the underlying logic isn't complex: the industry has moved from a building phase to an execution phase, over 100 chains need integration, institutional clients bring entirely different compliance and security requirements, and DAO governance requires balancing stakeholders with diverse interests. This isn't traditional project management but coordination and judgment in an environment where rules are still being formed.
The industry appears to be shrinking on the surface, but core density is actually increasing. The 2018-2019 bear market similarly saw massive developer exodus, but afterward gave rise to paradigm-defining projects like Uniswap, Aave, and OpenSea that defined the 2020-2021 bull run. The builders remaining this round have more mature infrastructure, and the AI era offers them a stage even larger than the last cycle.
II. What Skills Have Those Who Stayed Brought With Them?
What special abilities has the Crypto industry truly cultivated in builders? To answer this, we must return to blockchain's underlying principles. Through cycles of bull and bear markets, this industry always runs on the same foundational rule: code is law, execution is final.
The 2016 DAO incident saw an attacker exploit a recursive call vulnerability to siphon off $36 million. The code had no bugs, logic executed exactly as intended, just with boundaries the designers hadn't anticipated. In 2021, the Poly Network cross-chain bridge was hacked, with $610 million transferred within hours. No platform could stop it, no institution could reverse it, no legal clause could seek restitution. This is a structural feature distinguishing crypto from almost all other industries: zero tolerance for error, almost no possibility for post-facto intervention.
This environment has forced the cultivation of a set of skills rarely required in other industries: building operational systems from scratch that strangers are willing to participate in, under conditions of missing rules and missing trust.
This ability has two layers. First, building trust from zero, relying on no external authority, only code and mechanisms to make strangers willing to place real assets into the system. Second, making judgments under dual technological and economic uncertainty, designing operational systems without regulatory frameworks, historical data, or industry standards to reference.
Both layers have concrete validation in crypto. Uniswap has no company guarantee, no KYC, no customer service. Anyone depositing funds into liquidity pools relies solely on trust in a few hundred lines of code and a set of economic mechanisms, achieving tens of billions in daily trading volume. MakerDAO has no central bank backing, no deposit insurance, maintaining DAI stability purely through on-chain governance and collateral mechanisms. During DeFi Summer it was more extreme: no regulatory framework, no audit standards, no historical data to reference; builders designed AMMs, lending protocols, liquidity mining—from concept to tens of billions in TVL in just months. This ability manifests differently in protocol-layer, application-layer, and governance-layer builders, but the underlying principle is the same.
The AI era is creating structurally highly similar problems. Model decision processes are opaque, outputs cannot be independently verified. AI agents are beginning to autonomously execute transactions and allocate funds, with accompanying rule systems and constraint mechanisms still absent. Large model companies control both the model and evaluation standards, leaving users lacking effective verification means. Computing power is highly concentrated among a few top tech giants, leading to monopoly pricing during demand surges. These issues point to the same core: the trust problem of autonomous systems is replaying at AI's larger scale.
Crypto builders have dealt with such issues for years in environments lacking external authoritative rules, only previously the scenario was on-chain protocols, now it's AI. And already a group has directly taken crypto-accumulated abilities into AI, achieving results.
III. How Are These Skills Being Repriced in the AI Era?
Cases transitioning from crypto to AI have become common in recent years, but upon examination, what they take differs.
The most direct path is the direct transfer of hardware and experience. CoreWeave's three founders Michael Intrator, Brian Venturo, and Brannin McBee started mining Ethereum with GPUs in 2017, scaling from one machine to thousands, closed their mining operations in 2022, with ChatGPT launching two months later, their GPUs directly becoming AI compute supply. They IPO'd on NASDAQ in March 2025 with a valuation of about $23 billion, later peaking near $70 billion.
OpenSea co-founder Alex Atallah dealt with aggregating and routing highly heterogeneous assets in the NFT market, applying the same experience to AI model routing, founding OpenRouter, serving over 5 million developers within two years, valued at $500 million.
Another type of migration is more noteworthy. NEAR founder Ilia Polosukhin is a co-author of the Transformer paper. After leaving Google, his initial aim was to build AI applications using natural language, but encountered a practical problem during development: making cross-border payments to data labelers worldwide, many of whom lacked bank accounts, making blockchain technology the optimal solution.
NEAR is now pivoting to an AI infrastructure platform, focusing on user-owned AI and Decentralized Confidential Machine Learning (DCML), allowing users to use AI services without exposing data. The decentralized architecture experience accumulated at NEAR became the hardest-to-replicate starting point in this direction.
Circle co-founder Sean Neville left to found Catena Labs, positioning it as AI-native banking, directly applying understanding of stablecoin infrastructure to AI agent financial scenarios, with a16z crypto leading an $18 million seed round. Aave and Lens Protocol veteran developer Nader Dabit moved to Cognition, bringing developer ecosystem building experience from multiple crypto protocols into the AI agent tools space.
These individuals didn't just take GPU hardware or user networks, but also intuition for mechanism design, experience building developer ecosystems, and judgment to build trustworthy systems from scratch when rules are absent. These abilities precisely correspond to three structural gaps encountered in AI scaling.
Compute Aggregation and Optimization
Compute is the most direct bottleneck for AI scaling. Training and inference require massive GPUs, demand fluctuates greatly, cloud providers are expensive with queues, and companies don't want to hoard hardware themselves. This problem has two layers: how to aggregate and allocate compute, and how to use aggregated compute more efficiently. Crypto builders have directly transferable accumulation on both layers.
Hyperbolic solves the distribution and trust problem. Founder Jasper Zhang brought decentralized mechanism design into the AI compute track: tokens incentivize dispersed GPU owners to contribute idle compute, but the core problem is trust.
Why believe computation results from a stranger's node are correct? The core innovation, PoSP, uses random sampling plus game theory, making honesty the dominant strategy for nodes, no need for full verification, low overhead, scalable, and reliable results. This mechanism is directly migrated from crypto's logic of verifying stranger node behavior.
MoonMath solves the efficiency problem. Its predecessor, Ingonyama, focused on ZK hardware acceleration, boosting ZK proof generation speed severalfold under extreme computational constraints. Now pivoting to Physical AI performance layer, working on sparse attention acceleration for video diffusion models (LiteAttention), low-rank decomposition for FFN layers (LiteLinear), and training backpropagation acceleration (BackLite). From ZK acceleration to AI inference acceleration, the underlying ability is the same: making math run faster under extreme computational constraints. The track changed, but the accumulation wasn't wasted.
AI Governance and Incentive Mechanism Design
When multiple AI agents begin collaborating on tasks, how to ensure they don't compromise the overall system while pursuing individual goals. Each participant pursues its own objective function, with no guarantee the system functions normally when combined, and agents operate far faster than human intervention windows.
This is a problem type crypto builders have repeatedly tackled in DAO governance and tokenomics design: making participants with completely different interests operate according to the system's preset direction without a central authority. Crypto's answer is economic mechanisms: non-compliant operations incur real economic costs, rules are written in code, automatically executed.
EigenLayer directly migrates this mechanism to AI scenarios. Through its restaking mechanism, nodes must stake assets before participating in collaboration; non-performance or violations trigger automatic penalties; rules are not suggestions but rigid boundaries with real economic consequences. EigenCloud extends this logic to AI agent verifiable computation and collaborative governance, forcing agents pursuing their own goals to stay within preset boundaries. Constraining agents with economic mechanisms is far more reliable than constraining them with ethical guidelines.
AI Agent Autonomous Payments
An even more foundational problem: how do agents pay? Traditional payment systems are designed for humans: credit cards require accounts, bank transfers require authorization, each step assuming a human operator with identity and patience. Agents don't wait; they may initiate numerous requests per second, each potentially involving micro-payments; traditional payment pipelines fail directly in this scenario.
Stablecoins and on-chain rules are infrastructure crypto builders have already built, natively supporting programmability, no authorization, and 24/7 operation. These three characteristics happen to be hard requirements for agent payment scenarios, lacking only a protocol layer connecting stablecoins to agent workflows.
x402, launched by Coinbase in May 2025, activates the HTTP 402 status code, embedding stablecoin payments directly into HTTP requests; agents complete payment while initiating requests, no account needed, settlement in about two seconds. As of April 2026, the x402 protocol has processed over 165 million transactions, cumulative volume ~$50 million, with 69,000 live agents (data source: x402 Foundation); Cloudflare, AWS, Stripe, and Anthropic MCP have all integrated. Agent payments are already a track with real traffic.
The three directions correspond to three structural gaps encountered in AI scaling: compute aggregation and efficiency, incentive alignment for multi-agent collaboration, and autonomous payment infrastructure. These problems have no ready answers in traditional software architecture, but crypto has corresponding experience. The abilities haven't vanished, just found new carriers.
IV. The New Role of Builders: From Smart Contract Coders to Rule Makers for AI
AI scaling is creating a previously non-existent functional gap. Not a gap for technical talent, but for people capable of designing trust mechanisms in autonomous systems. As the service target shifts from humans to AI, the role of crypto builders is being redefined.
The table below compares changes across specific functional paradigm dimensions:
The core difference between the two paradigms isn't the tech stack, but the way trust is established and rules are executed. In the Pre-AI era, crypto builders faced human participants, rules were written into contracts, tolerance for error was zero, but system boundaries were relatively clear.
In the AI-Native era, when the interaction target becomes autonomously operating AI agents, the problem to solve is: agent behavior is unpredictable, execution speed far exceeds human intervention windows, and system boundaries themselves need redefinition under greater uncertainty. The functional positioning of crypto builders is shifting from "writing secure contracts" to "designing trustworthy mechanisms for AI autonomous systems."
Hiring by leading institutions already reflects this change:
▲ Core AI/Data positions actively opened by leading exchanges in Q1 2026
Source: Gate Research Institute
Hiring by leading exchanges and institutions in 2026 clearly reflects this trend: no longer simply hiring AI engineers or crypto developers, but seeking people who can connect both sides—those who understand on-chain incentive distortions and governance games, can deeply embed AI tools into crypto workflows, and design mechanisms aligning agents with regulation and users long-term.
Capital allocation directions already reflect this judgment. Paradigm is raising a new fund up to $1.5 billion, expanding investment scope from crypto to AI and robotics. Haun Ventures completed a $1 billion Fund II, focusing on financial infrastructure where crypto and AI converge, especially payments, stablecoins, and agent-to-agent economic systems supporting AI agent autonomous trading and coordination.
a16z crypto raised a $2.2 billion fifth fund (Crypto Fund V), explicitly stating 100% will be allocated to crypto. Facing the complexity and opacity of the AI era, they will focus on applying crypto's transparency, verifiability, and decentralization characteristics. According to PitchBook data, in 2025 US crypto VC investment, about 40% of capital flowed to companies also involved in AI business, significantly up from 2024.
Similarly crypto builders transitioning to AI, but chosen paths differ markedly across market environments.
In the US, as the regulatory environment clarified relatively, protocol-layer innovation gained real survival space. Capital network density is high, the path from idea to funding is short, with larger error tolerance. Hyperbolic, EigenCloud, Gensyn, Ritual—these projects share the characteristic of designing new mechanisms from scratch, not simple application integration on existing systems. Top VCs have clear investment theses for directions like "verifiable computation, Agent coordination, decentralized ML," willing to provide ample tolerance for early technical exploration.
Asia's situation differs. Singapore and Hong Kong more often assume roles in compliant implementation and institutional capital transit, with relatively conservative regulatory frameworks, lower tolerance for pure protocol-layer innovation. Crypto-background builders transitioning to AI more often choose application-layer and industry integration paths—leveraging crypto-accumulated user bases, payment capabilities, or data assets to quickly integrate into AI products and services.
This isn't a capability gap, but a path selection difference caused by varying market signals and regulatory environments: the US encourages underlying mechanism innovation and early technical exploration, Asia emphasizes compliance-friendliness, rapid monetization, and deep integration with traditional industries.
Returning to the opening GitHub curve. Monthly active developers dropped from 45K to 23K, superficially suggesting industry contraction. But among those remaining, the proportion of established devs hit a historical high, flocking to ecosystems with real users, while being repriced by the AI industry in unprecedented ways.
As AI scaling encounters structural bottlenecks in compute aggregation, agent autonomous payments, data and decision verifiability, and privacy coordination, these builders stand at the intersection of Crypto and AI. The long-accumulated sensitivity to rules, incentives, and truthfulness is gradually transforming into system-level abilities scarce in the AI era.
As an investment institution deeply involved in crypto infrastructure since 2017, IOSG's judgment on this line isn't limited to observation. We participated in EigenLayer's investment when its restaking mechanism wasn't widely recognized, led the seed round for Ingonyama (now MoonMath) betting on ZK hardware acceleration migrating to the AI performance layer, and invested in Hyperbolic in 2024, optimistic about its path using crypto-native verification mechanisms to solve decentralized compute trust problems.
The common logic behind these moves is: the trust, coordination, and verification problems encountered during AI scaling will ultimately require the mechanism design abilities accumulated in the crypto industry. We believe the convergence of crypto and AI isn't a narrative, but a structural opportunity already unfolding.
