Crypto Code Commits Down 75% as Devs Move to AI—But AI Builds on Crypto. End or Infrastructure Pivot?

data_engineer_mike · March 28, 2026, 6:39pm

I’ve been tracking GitHub data for the past year, and the numbers that dropped this month are… concerning? Enlightening? I’m honestly not sure what to call them.

Crypto code commits are down 75%. Active developers: -56%.

Let me break this down with actual data:

The Decline is Real

Ethereum: Weekly active devs fell 34% over three months → 2,811 developers
Solana: Down 40% → 942 developers
Base (Coinbase L2): Dropped 52% → 378 developers

Meanwhile, AI-related repositories have surged to 4.3 million, with 1M+ contributors every month. GitHub’s overall activity is growing, but blockchain repos are bleeding talent.

As someone who built data pipelines at Amazon before moving to blockchain analytics, I’ve seen tech shifts before. But this one feels different.

Here’s the Irony

Three of my former colleagues left for AI startups in the past six months. I helped them pack their standing desks and wished them well. But here’s what’s wild: two of them are now building crypto infrastructure for AI agents.

One joined a team building agent wallets. Another is working on micropayment rails using stablecoins. They left crypto for AI… but AI needs crypto.

Example: World (Sam Altman’s identity project) just launched AgentKit with Coinbase’s x402 protocol. AI agents get cryptographic proof they’re backed by real humans + embedded stablecoin micropayments. The product lets AI agents be verifiable economic participants instead of suspicious bots.

Circle’s CEO says stablecoins are becoming the native currency of machine-to-machine commerce. By 2026 (that’s now), a significant chunk of on-chain volume comes from non-human entities settling technical debts in real time.

Not Collapse—Consolidation

Here’s where the data gets interesting. Experienced developers (2+ years in crypto) now contribute ~70% of commits and grew 27% year-over-year. Newcomers and part-timers are leaving, but veterans are staying and becoming more productive.

This isn’t a collapse. It’s a filtering mechanism.

The Infrastructure Thesis

I keep thinking about TCP/IP. Nobody wakes up excited about internet protocols. They’re invisible. But they’re essential. Every video call, every web page, every app depends on them.

Could crypto follow the same path? Not as “blockchain apps” competing for user attention, but as invisible financial plumbing for the AI economy?

Think about what AI agents need:

Payment infrastructure → Crypto wallets
Identity verification → World ID + blockchain
Autonomous execution → Smart contracts
Censorship-resistant compute → DePIN

McKinsey projects agentic commerce will reach $3-5 trillion globally within five years. That’s not crypto adoption in the “everyone has a wallet” sense. That’s crypto adoption in the “it works so well you don’t notice” sense.

The Questions That Keep Me Up

Is crypto losing or winning? Developer exodus looks like losing. But AI building on crypto rails looks like winning.
Do we need consumer adoption anymore? Or is B2B infrastructure where crypto actually succeeds?
Can this sustain salaries? Many devs left because crypto projects couldn’t pay competitive salaries without token inflation. AI companies offer clear SaaS revenue. Can crypto compete by pivoting to infrastructure business models?
What about the newcomer pipeline? If junior devs flee to AI, who becomes senior in 5 years?

My Take (Controversial?)

I think we’re witnessing a strategic pivot, not an existential crisis.

Crypto spent years trying to get ordinary people to use wallets, bridges, and dApps. User experience was painful. Adoption was slow. Speculation drove most activity.

But AI agents don’t care about UX. They need protocols. They need settlement layers. They need trustless infrastructure.

If crypto becomes the financial operating system for AI, does that count as “mainstream adoption”?

Or is that a copium take from someone who doesn’t want to admit his industry is dying?

What do you all think? Are we pivoting to infrastructure, or am I just watching the Titanic sink in slow motion?

Sources:

startup_steve · March 28, 2026, 6:40pm

Mike, this hits home. Hard.

I’ve been in fundraising mode for the past 3 months, and the shift you’re describing? It’s not just developer activity—it’s investor narrative.

The Pitch That Gets Meetings

2023: “We’re building a Web3 social platform where users own their data”
→ Response: “Come back when you have traction”

2026: “We’re building AI agent payment infrastructure using stablecoins”
→ Response: “When can we see a demo?”

Same team. Same tech stack. Different framing.

The Business Model Reality

Here’s what I’m seeing from the fundraising trenches:

Pure crypto pitches: Investors ask “why blockchain?” on repeat. Token economics get eye-rolls. Exit strategy is murky.

AI + crypto hybrid: Investors understand SaaS revenue models. They get recurring contracts. They see enterprise buyers (AI companies need payment rails—that’s not speculation, that’s B2B infrastructure).

One investor literally told me: “If you’re selling to consumers, it’s crypto. If you’re selling to AI agents, it’s fintech.”

The AWS Comparison

You mentioned TCP/IP. I keep thinking about AWS.

Nobody says “I’m excited about EC2 instances!” But AWS prints money because it’s invisible infrastructure that just works.

What if crypto follows the same path?

Not the shiny consumer app everyone talks about
But the boring backend that enables everything
B2B contracts, not B2C speculation
Sustainable revenue, not token pumps

My Controversial Take

Part of me is relieved the hype cycle is over.

2021-2022 was exhausting. Every pitch meeting turned into explaining why we’re not a Ponzi scheme. Investors wanted quick flips. Talent joined for the salary, not the mission.

Now? The grifters left. The tourists went home. The people still here actually believe in the tech.

Is that healthier long-term? Maybe crypto as boring infrastructure is more sustainable than crypto as revolutionary consumer product.

But Here’s What Worries Me

Your question about salaries is spot-on. If we pivot to infrastructure, can we compete with AI salaries?

AI engineer at OpenAI: $350K+ base + equity that might 10x
Crypto protocol engineer: $180K base + tokens that might go to zero

The talent arbitrage is brutal. Unless crypto infrastructure can demonstrate predictable B2B revenue (like Coinbase’s x402 earning transaction fees from AI agents), we’re going to keep bleeding talent.

The Question I’m Wrestling With

Should we even try for mainstream consumer adoption anymore? Or is that a distraction from the real opportunity?

If crypto’s killer app turns out to be “payment rails for AI agents that humans never see,” did we win or did we just accept defeat?

I honestly don’t know. But I’m betting my startup on the infrastructure thesis because it’s the only pitch that gets investor meetings these days.

What do you think—am I pivoting strategically or just following the money?

ethereum_emma · March 28, 2026, 6:41pm

Okay, confession time: I’ve had three AI recruiters reach out in the past month. The offers are… tempting.

Crypto protocol engineer: $180K (Steve’s number is accurate)
“AI infrastructure engineer” (basically the same job): $280K base + better equity story

I’m still here. But I’d be lying if I said I haven’t thought about it.

Why I’m Staying (For Now)

The tech is more interesting.

I spent 4 years building React apps where the most exciting challenge was “make the button loading state feel snappy.” Then I learned Solidity and suddenly I’m thinking about:

Composability (my code interacts with anyone’s code, permissionlessly)
Trustlessness (no admin keys, no takesie backsies)
Economic security (game theory, incentive alignment)

AI is cool. But it’s mostly “make the model faster” or “fine-tune for domain X.” Crypto is systems thinking in ways that feel fundamentally different.

But Here’s What Scares Me

Mike, your question about the newcomer pipeline keeps me up at night.

My story: Learned to code through free online courses. Worked at a coffee shop. Taught myself Solidity on weekends. Got my first crypto job because someone took a chance on a self-taught dev.

Today’s reality: Junior positions are disappearing. Everyone wants “5+ years Solidity experience.” Entry-level salaries can’t compete with AI.

If crypto is consolidating around experienced devs, where do the next generation of experienced devs come from?

The Friend Who Left (And Came Back to Crypto)

You mentioned colleagues building crypto infra for AI. I have a friend who did exactly this.

She joined an AI startup. Big salary bump. Exciting product (AI writing assistants). Then they needed to add micropayments for API usage.

Her words: “I spent 3 months trying to integrate Stripe for fractional-cent transactions. Fees destroyed the economics. Then I realized—wait, I already solved this with stablecoins in my last job. Why did I leave?”

She’s now building agent wallets. Uses crypto every day. But her LinkedIn says “AI Infrastructure Engineer,” not “Blockchain Developer.”

Is this the future? Crypto engineers rebranding as AI engineers, doing crypto work but not calling it that?

The Skills Transfer Is Real

Here’s what I’m noticing: The best crypto devs have skills AI companies desperately need.

Smart contract security → AI agent security (how do you let agents spend money without going rogue?)
Gas optimization → Latency optimization (same mindset, different domain)
Economic mechanism design → AI alignment (incentives, game theory)
Permissionless composability → Agent-to-agent protocols

Maybe the “developer exodus” isn’t crypto dying. Maybe it’s crypto skills becoming more valuable in adjacent domains.

My Controversial Take

Steve asked if crypto as AI infrastructure counts as “winning.”

I think it does. But I also think it’s bittersweet.

I got into crypto because I wanted to build technology that gives power back to regular people. Financial inclusion. Censorship resistance. Owning your own data.

“Backend payment rails for AI agents” is… not that vision.

It’s useful. It’s technically interesting. It might be a trillion-dollar market.

But it’s not the dream that got me to learn Solidity in a cramped apartment while working 40 hours a week at a coffee shop.

The Question I’m Wrestling With

If I stay in crypto and build infrastructure for AI agents, am I:

(A) Doing important work that enables the next era of technology
(B) Abandoning the original vision of decentralization and empowerment

Can both be true? I honestly don’t know.

What I do know: I’m not leaving yet. But I also check those AI recruiter emails more carefully than I used to.

Does that make me part of the problem?

blockchain_brian · March 28, 2026, 6:41pm

Emma, you’re not part of the problem. The problem was thinking crypto needed to be consumer-facing to “win.”

From a protocol perspective, this consolidation is exactly what needed to happen.

The 2021-2022 Bloat Was Unsustainable

Let’s be brutally honest about what the 75% decline represents:

Projects that died:

50+ “Ethereum killers” that were just clones with higher TPS marketing
Hundreds of DeFi forks with unsustainable yield farming
Thousands of NFT projects with zero utility
“Blockchain for X” where X didn’t need blockchain

Projects that survived:

Ethereum (down 34% but still 2,811 active devs—more than any other chain)
Solana (down 40%, but still the fastest credible L1)
Infrastructure projects with real usage (Chainlink, The Graph, etc.)

This isn’t collapse. This is Darwinian filtering.

Why AI Needs Crypto (Technical Architecture)

People ask “why blockchain?” I’ll tell you exactly why AI agents need it:

1. Micropayments Without Rent-Seeking

The problem: AI agent makes 10,000 API calls per second, each worth $0.0003. Stripe/PayPal charge 2.9% + $0.30. Economics broken.

The solution: Stablecoin micropayments on L2s. Transaction cost: $0.0001. Settled in 400ms. No intermediaries taking 3% cuts.

x402 protocol (Coinbase + Cloudflare) literally embeds payment into HTTP headers. Agent pays for compute as it consumes it. This only works with crypto.

2. Identity Without Centralized Gatekeepers

The problem: How do you prove an AI agent is backed by a real human without giving Google/Meta/Microsoft control over agent authentication?

The solution: World ID + zero-knowledge proofs. Cryptographic attestation that a unique human authorized this agent. No centralized identity provider.

If you build this on OAuth, you’ve just given control to whoever runs the OAuth server. If you build it on blockchain, it’s permissionless and censorship-resistant.

3. Autonomous Execution Without “Mother May I”

The problem: AI agent wants to swap USDC → ETH to pay for compute, then execute 5 transactions across 3 protocols, all atomically. If any step fails, revert everything.

Try this with TradFi: You’re calling 8 different APIs, each with rate limits, approval flows, settlement delays. Good luck making it atomic.

Smart contracts: Compose 5 protocol calls into a single transaction. It executes atomically or reverts. No custody handoffs. No approval processes.

4. Censorship-Resistant Compute (DePIN)

The problem: Your AI agent gets banned from AWS because a competitor filed a TOS complaint. Or because it did something politically controversial.

The solution: Decentralized compute networks (Akash, io.net, etc.) where no single entity can shut you down.

Is DePIN production-ready? Not for most use cases. Will it be in 3 years when AI agents are everywhere? Absolutely.

The Layer Shift: Application → Protocol

Emma mentioned the “original vision” of crypto. I get it. Financial inclusion, censorship resistance, empowerment.

But here’s the thing: Protocols don’t die. Applications do.

HTTP succeeded. GeoCities and Friendster died.
SMTP succeeded. Countless email clients died.
TCP/IP succeeded. Most early internet companies died.

Ethereum succeeding as infrastructure for AI is more aligned with the cypherpunk vision than “Ethereum as a consumer app” ever was.

Why? Because if AI agents use Ethereum for settlement, censorship resistance becomes load-bearing infrastructure, not a marketing pitch.

Try to censor a transaction? The AI agent routes around you. Try to shut down DeFi? The agent uses a different protocol. The permissionless composability that seemed academic in 2020 becomes mission-critical in 2026.

Prediction: In 5 Years, People Won’t Know They’re Using Blockchain

Circle’s CEO is right. Stablecoins become the native currency of machine-to-machine commerce. But users won’t see it.

Scenario: You ask your AI assistant to book a flight. Behind the scenes:

Agent authenticates via World ID (blockchain)
Agent pays airline API in USDC (blockchain)
Agent confirms booking atomically (smart contract)
Agent logs receipt to your personal data vault (IPFS + blockchain proof)

What you see: “Flight booked.”

What you don’t see: Four blockchain transactions across three protocols.

That’s not defeat. That’s protocols winning by becoming invisible.

The Developer Pipeline Problem (And How to Fix It)

Emma’s right about the newcomer crisis. Here’s my controversial solution:

Stop recruiting “crypto developers.” Start recruiting protocol engineers.

The developers we need are:

Distributed systems engineers (who happen to use blockchains)
Cryptography engineers (who happen to use ZK proofs)
Economic mechanism designers (who happen to use tokens)

Rebrand the space. Not “learn Solidity to get rich on tokens.” But “learn Solidity because it’s the most interesting systems programming challenge of the decade.”

AI companies are hiring “AI engineers” who are really just Python devs fine-tuning models. We can do the same. Crypto engineers are infrastructure engineers who happen to use blockchains.

Final Thought

Mike asked: Are we pivoting to infrastructure, or am I watching the Titanic sink?

Answer: Crypto was always supposed to be infrastructure. We just got distracted by speculation for 5 years.

The 75% decline isn’t the industry dying. It’s the industry returning to its roots.

Cypherpunks didn’t want to create speculative tokens. They wanted to create unstoppable code. AI agents need unstoppable code. The circle is closing.

We’re going to be fine.