# pzzaworks

> Baked with code, served hot. The kitchen behind Web3 and AI products.

pzzaworks is the personal site of Berke, a full-stack engineer building at the intersection of Web3, DeFi, and AI agents.

- Site: https://pzza.works
- Full content for LLMs: https://pzza.works/llms-full.txt

## Projects

- [Astreus](https://pzza.works/projects/astreus): Open-source AI agent framework for building autonomous systems that solve real-world tasks effectively.
- [Buidl Now](https://pzza.works/projects/buidl-now): A comprehensive suite of 109 developer tools and Web3 utilities for builders who ship fast.
- [Forgetless](https://pzza.works/projects/forgetless): Smart context optimization library for LLMs. Compress massive content to fit your token budget with intelligent prioritization.
- [Lunark](https://pzza.works/projects/lunark): AI companion platform that makes blockchain human-friendly through natural language interactions.
- [Titan](https://pzza.works/projects/titan): DeFi super app combining token swapping, liquid staking, overcollateralized lending, and on-chain governance with native Uniswap V4 integration.
- [Void DEX](https://pzza.works/projects/void-dex): Privacy-first DEX aggregator that combines best swap rates with Railgun's zero-knowledge privacy technology.

## Articles (The Oven)

- [Titan: Building a DeFi Super App](https://pzza.works/oven/titan-building-defi-super-app): A technical deep dive into Titan - a complete DeFi protocol featuring token swaps, concentrated liquidity, liquid staking with sTITAN, overcollateralized lending with tUSD, and snapshot-based on-chain governance.
- [Forgetless: Smart Context Optimization for LLMs](https://pzza.works/oven/forgetless-smart-context-optimization-llms): Every LLM has a context limit. When your documents, conversations, and knowledge exceed that limit, you're forced to make difficult choices about what to include. Forgetless makes those choices intelligently, achieving up to 14.5x compression while preserving what matters most.
- [DEX Aggregation in VoidDex: Building an Optimal Route Finder](https://pzza.works/oven/voiddex-dex-aggregation-routing): How VoidDex's backend finds the best swap routes across multiple DEXes, covering the quote service architecture, route optimization strategies, and fee calculation system.
- [VoidDex Router: Designing a Multi-DEX Swap Router with Split Routing](https://pzza.works/oven/voiddex-smart-contract-router-design): Deep dive into VoidDex's smart contract architecture, covering the router design, DEX adapter system, and split routing for optimal execution.
- [Privacy Architecture in VoidDex: Integrating Railgun's zkSNARK System](https://pzza.works/oven/voiddex-privacy-architecture-railgun): How VoidDex implements private token swaps using Railgun's zero-knowledge proof system, covering the shield/unshield flow, client-side proof generation, and the Waku P2P broadcaster network.
- [Building VoidDex: A Technical Deep Dive into Privacy-Preserving DEX Aggregation](https://pzza.works/oven/building-voiddex-privacy-preserving-dex): Every swap you make on a decentralized exchange is public. VoidDex is my attempt at building a DEX aggregator that maintains the benefits of decentralized trading while adding a cryptographic privacy layer using zkSNARK proofs.
- [Broadcasters & The RAILGUN Ecosystem: Gasless Private Transactions](https://pzza.works/oven/broadcasters-railgun-ecosystem-gasless-transactions): We've covered how RAILGUN provides privacy through shielding, how zero-knowledge proofs make it possible, and how PPOI maintains compliance. But there's a practical problem we haven't addressed: how do you submit a private transaction to the blockchain without revealing who you are?
- [Private Proofs of Innocence: Compliance Without Compromising Privacy](https://pzza.works/oven/private-proofs-of-innocence-compliance-privacy): Privacy protocols face a fundamental tension. On one hand, users deserve financial privacy. On the other hand, bad actors would love to use privacy tools to hide illicit funds. RAILGUN's answer is Private Proofs of Innocence, or PPOI.
- [Zero-Knowledge Proofs Explained: The Cryptography Behind RAILGUN](https://pzza.works/oven/zero-knowledge-proofs-cryptography-behind-railgun): We've covered what RAILGUN does and how to use it. Now let's explore the magic underneath: zero-knowledge proofs. This is the cryptographic foundation that makes private transactions possible without requiring trust in any central party.
- [Shield & Unshield: How RAILGUN Protects Your Tokens](https://pzza.works/oven/shield-and-unshield-how-railgun-protects-tokens): In the first article, we covered why on-chain privacy matters and how RAILGUN brings it to DeFi. Now let's get practical. How do you actually make your tokens private? What happens under the hood when you shield and unshield?
- [What is RAILGUN? The Privacy Layer DeFi Has Been Waiting For](https://pzza.works/oven/what-is-railgun-privacy-layer-defi): If you've spent any time in DeFi, you know the feeling. Every swap you make, every token you hold, every wallet you interact with, it's all out there. Public. Permanent. RAILGUN is a smart contract system that brings zero-knowledge privacy directly to EVM chains.
- [The Future of Privacy: 2026 and Beyond](https://pzza.works/oven/future-of-privacy-2026-beyond): We're at a pivotal moment in blockchain history. After years of research, failed experiments, and regulatory battles, privacy is finally becoming real. This isn't another hype cycle. The numbers, the institutional adoption, the technology maturity - everything is aligning for privacy to become blockchain's default, not its exception.
- [Real-World Privacy Applications](https://pzza.works/oven/real-world-privacy-applications): We've covered the technology, the roadmap, and the tools. Now it's time for the question that matters most: What can you actually DO with blockchain privacy? The answer might surprise you - privacy isn't just for hiding transactions. It's unlocking entirely new categories of applications.
- [Kohaku: Ethereum's Privacy-First Wallet Revolution](https://pzza.works/oven/kohaku-wallet-revolution): When Vitalik Buterin took the stage at Devconnect in Buenos Aires in November 2025, he didn't mince words about Ethereum's state. Kohaku is Ethereum's answer - a privacy-first wallet framework designed to make zero-knowledge proofs accessible to everyday users.
- [Ethereum 2025-2026 Privacy Roadmap: The Dawn of Private Infrastructure](https://pzza.works/oven/ethereum-privacy-roadmap-2025-2026): For years, privacy on Ethereum was an afterthought, something left to third-party protocols and layer-2 solutions. That changed in September 2025. The Ethereum Foundation made privacy an official part of the roadmap through the Privacy Stewards of Ethereum (PSE) initiative.
- [zkSNARK vs zkSTARK: Ethereum's Two Privacy Champions](https://pzza.works/oven/zksnark-vs-zkstark): In our previous article, we explored how zero-knowledge proofs are solving blockchain's privacy paradox. But not all ZK-proofs are created equal. Two technologies are competing to become the standard for Ethereum's privacy future: zkSNARKs and zkSTARKs.
- [Blockchain's Privacy Paradox and the Solution: Zero-Knowledge Proofs](https://pzza.works/oven/blockchain-privacy-paradox): For the past 10 years, blockchain has operated with radical transparency - every transaction is publicly recorded, which leaves users vulnerable to phishing attacks. While transparency is blockchain's greatest strength for trust and verification, it's also becoming its Achilles heel for mainstream adoption.
- [Advanced Memory Systems for AI Agents: Engineering for Persistence and Contextual Intelligence](https://pzza.works/oven/advanced-memory-systems-ai-agents): Advanced memory systems are a critical component that enables truly intelligent agent behavior. Drawing from experience implementing production-grade AI agents, this article shares technical patterns for designing memory architectures that dramatically enhance agent capabilities.
- [Orchestrating Multi-Agent Systems: Technical Patterns for Complex Problem Solving](https://pzza.works/oven/orchestrating-multi-agent-systems): The true power of AI agents emerges when they work together in orchestrated systems. While individual agents offer impressive capabilities, their potential multiplies exponentially when organized into coordinated multi-agent architectures. This article explores the technical patterns that enable effective collaboration between specialized AI agents.
- [The Technical Architecture of Modern AI Agents](https://pzza.works/oven/technical-architecture-modern-ai-agents): This is the first in my technical series on AI agent architecture. Having built numerous production-grade AI agent systems, I'm sharing insights from my implementation experience. I'll take you behind the scenes into the technical foundations that make these systems work.
- [Understanding AI Agents: Architecture, Implementation, and Future Directions](https://pzza.works/oven/understanding-ai-agents-architecture-implementation): As an AI agent engineer who has worked extensively with frameworks like LangChain, ElizaOS and built custom framework for Lunark AI, I've seen firsthand how AI agents are transforming the way we think about software. In this article, I'll explain what AI agents are, how they work under the hood, and where I see this technology heading.
- [Understanding Paymasters: Building Gas Abstraction on zkSync](https://pzza.works/oven/understanding-paymasters-gas-abstraction-zksync): As a blockchain developer who has built multiple Paymasters and implemented zkSync's native account abstraction, I've seen firsthand how these technologies are revolutionizing blockchain UX. In this post, I'll dive deep into Paymasters, their implementation on zkSync, and why they're a game-changer for Web3 adoption.
- [Multi-Chain DEX Aggregation: How Lunark Finds the Best Swap Rates](https://pzza.works/oven/multi-chain-dex-aggregation-lunark): A technical walkthrough of Lunark's DEX aggregation system, covering quote fetching across Uniswap, SushiSwap, Curve, PancakeSwap, and TraderJoe, fee tier optimization, and the challenges of building a multi-protocol swap system.
- [Real-Time Streaming with Socket.IO: Building Responsive Web3 Chat](https://pzza.works/oven/real-time-streaming-socketio-web3-chat): How Lunark implements real-time message streaming using Socket.IO, including the challenges of React state batching, room-based broadcasting, and coordinating blockchain transactions with chat responses.
- [Building Lunark: AI Agent Architecture for Blockchain Operations](https://pzza.works/oven/building-lunark-ai-agent-blockchain): A deep dive into how Lunark uses the Astreus framework to create an AI agent that can execute blockchain operations through natural language. Covers the tool system, agent caching, context injection, and the challenges of building autonomous blockchain agents.
- [zkTLS: Bridging Web2 Privacy with Web3 Verification](https://pzza.works/oven/zktls-bridging-web2-privacy-web3-verification): I've built several dApps and AI agents using Zero-Knowledge Transport Layer Security (zkTLS) and wanted to share why this technology is so transformative. zkTLS represents a significant advancement in blockchain security architecture, combining zero-knowledge cryptography with enhanced TLS protocols.

## Notes (Bites)

- [The Trilemma Still Holds](https://pzza.works/bites/the-trilemma-still-holds): Decentralization, security, scalability - pick two. Years of research shifted the tradeoffs but didn't eliminate them.
- [The Oracle Problem Persists](https://pzza.works/bites/the-oracle-problem-persists): Smart contracts can't see the outside world. Oracles bridge that gap but the trust assumptions never fully disappear.
- [MEV Evolution](https://pzza.works/bites/maximal-extractable-value-evolution): MEV went from obscure research topic to billions in annual extraction. The evolution shaped how we think about blockchain fairness.
- [The Sequencer Centralization Question](https://pzza.works/bites/the-sequencer-centralization-question): Most rollups run single sequencers. The decentralization roadmaps exist but execution is slow.
- [State Channels vs Rollups](https://pzza.works/bites/state-channels-vs-rollups): State channels came first but rollups won. Understanding why helps predict what wins next.
- [Finality Is a Spectrum](https://pzza.works/bites/finality-is-a-spectrum): Nothing is truly final immediately. Understanding the different finality guarantees helps make better tradeoffs.
- [Post-Quantum Crypto Timeline](https://pzza.works/bites/post-quantum-crypto-timeline): Quantum computers will break current cryptography eventually. The migration needs to start before they arrive.
- [Economic vs Cryptographic Security](https://pzza.works/bites/economic-vs-cryptographic-security): Cryptography is binary - broken or not. Economic security degrades gracefully. Understanding the difference matters for system design.
- [BLS Signatures Everywhere](https://pzza.works/bites/bls-signatures-everywhere): Aggregate thousands of signatures into one. BLS made Ethereum's consensus possible and enables much more.
- [SNARK vs STARK Tradeoffs](https://pzza.works/bites/snark-vs-stark-tradeoffs): Both prove computation correctly happened. The tradeoffs between them shape which rollups use what.
- [Agent Trust Frameworks](https://pzza.works/bites/agent-trust-frameworks): How do you trust an agent you've never interacted with? Reputation, staking, and attestation systems emerging.
- [Orderbook DEXs Comeback](https://pzza.works/bites/orderbook-dexs-comeback): On-chain orderbooks became viable with L2 scaling. AMMs aren't the only game in town anymore.
- [Prediction Markets Went Mainstream](https://pzza.works/bites/prediction-markets-went-mainstream): Prediction markets broke into mainstream consciousness. Election markets got more attention than polls.
- [Cross-chain Identity Forming](https://pzza.works/bites/cross-chain-identity-forming): Your on-chain identity spanning multiple chains. Unified reputation that follows you across ecosystems.
- [MEV Redistribution Experiments](https://pzza.works/bites/mev-redistribution-experiments): Projects trying to return MEV to users instead of letting searchers capture everything. The results are mixed but promising.
- [Developer Experience Improved](https://pzza.works/bites/developer-experience-improved): Building on crypto got significantly easier. Better tools, better docs, better debugging. The rough edges smoothed.
- [The DAO Treasury Problem](https://pzza.works/bites/dao-treasury-problem): Billions in DAO treasuries sitting idle or poorly managed. Governance struggles to deploy capital effectively.
- [Social Tokens Found Their Niche](https://pzza.works/bites/social-tokens-found-their-niche): Social tokens survived the hype cycle. Small communities using them for access and coordination.
- [Yield Farming Grew Up](https://pzza.works/bites/yield-farming-grew-up): DeFi Summer's degen farms evolved into sustainable yield infrastructure. Real yields from real economic activity.
- [Hardware Wallets Evolved](https://pzza.works/bites/hardware-wallets-evolved): Hardware wallets added smart account support, biometrics, and wireless connectivity. The UX gap with software wallets closed.
- [Agent Security Lessons](https://pzza.works/bites/agent-security-lessons): Recent agent security incidents taught expensive lessons. Prompt injection, key management, access control all failed publicly.
- [DePIN Proved Itself](https://pzza.works/bites/depin-proved-itself): Decentralized physical infrastructure networks went from concept to real networks with real utility.
- [Onchain Games Shipping](https://pzza.works/bites/onchain-games-shipping): Fully on-chain games went from demos to real products. The UX got good enough for actual players.
- [Privacy Tech Renaissance](https://pzza.works/bites/privacy-tech-renaissance): ZK, MPC, FHE all maturing together. The privacy stack is becoming usable not just theoretically possible.
- [Execution Layer Diversity](https://pzza.works/bites/execution-layer-diversity): Different VMs for different use cases. The EVM isn't the only game in town anymore.
- [Governance Lessons Learned](https://pzza.works/bites/governance-lessons-learned): DAO governance evolved through trial and error. Some patterns emerged as clearly better than others.
- [Infrastructure That Scaled](https://pzza.works/bites/infrastructure-that-scaled): The boring infrastructure work of past years paid dividends. RPC providers, indexers, oracles all handled the load.
- [Agent Marketplaces Emerging](https://pzza.works/bites/agent-marketplaces-emerging): Buying and selling agent services on-chain. A new primitive for AI economics is forming.
- [NFT Utility Found Its Footing](https://pzza.works/bites/nft-utility-found-its-footing): Beyond JPEGs, NFTs became access passes, identity primitives, and composable game items.
- [DeFi Composability Got Complex](https://pzza.works/bites/defi-composability-got-complex): Money legos stacked too high. The interdependencies are hard to reason about now.
- [Gasless UX Is Standard](https://pzza.works/bites/gasless-ux-is-standard): Users don't think about gas anymore. Apps abstract it away or cover it entirely. The friction vanished.
- [Bridges Got Safer](https://pzza.works/bites/bridges-got-safer): After billions lost, bridge security improved dramatically. ZK proofs and light clients replaced trusted multisigs.
- [Validator Economics Shifting](https://pzza.works/bites/validator-economics-shifting): Staking yields compressed, MEV became crucial, restaking added complexity. Being a validator isn't simple anymore.
- [Protocol Owned Liquidity Evolved](https://pzza.works/bites/protocol-owned-liquidity-evolved): POL went from OlympusDAO experiment to standard treasury practice. Owning your liquidity beats renting it.
- [Account Abstraction Hit Mainstream](https://pzza.works/bites/account-abstraction-hit-mainstream): Smart accounts aren't experimental anymore. The majority of new wallets are contract-based now.
- [Light Clients Actually Work](https://pzza.works/bites/light-clients-actually-work): Verify blockchain state without downloading everything. Light clients graduated from theory to practice.
- [The Surge Is Happening](https://pzza.works/bites/the-surge-is-happening): Ethereum's scaling roadmap entering its most aggressive phase. The throughput improvements are finally visible.
- [Agent Communication Protocols](https://pzza.works/bites/agent-communication-protocols): Agents need to talk to each other. The protocols emerging are fascinating hybrid of HTTP and blockchain.
- [Cross-chain Messaging Standardized](https://pzza.works/bites/cross-chain-messaging-standardized): LayerZero, CCIP, Wormhole competing but also converging. Standards are emerging from the chaos.
- [Layer 3s Found Their Niche](https://pzza.works/bites/layer-3s-found-their-niche): L3s on L2s seemed like meme infrastructure. Turns out application-specific chains have real use cases.
- [Intents Are Eating DeFi](https://pzza.works/bites/intents-are-eating-defi): Express what you want, let solvers figure out how. Intent-based architecture is becoming the default.
- [Client Diversity Matters More](https://pzza.works/bites/client-diversity-matters-more): One bug in a dominant client could halt the network. Diversity isn't just nice to have, it's existential.
- [ZK Coprocessors Are Underrated](https://pzza.works/bites/zk-coprocessors-are-underrated): Offload heavy computation to ZK proving, bring results back to chain. Smart contracts just got way more powerful.
- [Stablecoin Regulations Clarified](https://pzza.works/bites/stablecoin-regulations-clarified): After years of uncertainty, stablecoin rules are taking shape. Some surprises, some expected, all consequential.
- [RWA Tokenization Hit Scale](https://pzza.works/bites/rwa-tokenization-hit-scale): Real world assets on-chain passed the experimental phase. Treasury bills, real estate, private credit all tokenized.
- [Uniswap v4 Hooks Changed Everything](https://pzza.works/bites/uniswap-v4-hooks-changed-everything): Custom logic at every stage of a swap. DEXs became programmable platforms overnight.
- [Bitcoin L2s Are Getting Serious](https://pzza.works/bites/bitcoin-l2s-are-getting-serious): BitVM, rollups on Bitcoin, programmability without changing the base layer. The Bitcoin L2 ecosystem exploded.
- [Agent Guardrails in Production](https://pzza.works/bites/agent-guardrails-in-production): Letting agents handle real money requires serious safety measures. The industry has learned some lessons the hard way.
- [Modular Thesis Is Winning](https://pzza.works/bites/modular-thesis-is-winning): Separate execution, settlement, consensus, data availability. The modular blockchain architecture proved itself.
- [MEV Supply Chain Matured](https://pzza.works/bites/mev-supply-chain-matured): Searchers, builders, relays, proposers. The MEV pipeline is now a sophisticated market with its own economics.
- [Shared Sequencers Unlock Composability](https://pzza.works/bites/shared-sequencers-unlock-composability): Multiple rollups sharing a sequencer enables atomic cross-rollup transactions. The fragmentation problem might have a solution.
- [Data Availability Wars Heat Up](https://pzza.works/bites/data-availability-wars-heat-up): Celestia, EigenDA, Avail all fighting for rollup DA. The modular thesis is playing out in real time.
- [Parallel EVM Is Real](https://pzza.works/bites/parallel-evm-is-real): Multiple transactions executing simultaneously. The EVM finally learns to use more than one core.
- [Based Rollups Feel Different](https://pzza.works/bites/based-rollups-feel-different): Letting L1 proposers sequence L2 transactions. Maximum decentralization at the cost of some flexibility.
- [Verkle Trees Are Coming](https://pzza.works/bites/verkle-trees-are-coming): Smaller proofs, stateless clients, portable state. Verkle trees could finally make Ethereum light clients viable.
- [Restaking Changed The Game](https://pzza.works/bites/restaking-changed-the-game): Using staked ETH to secure multiple protocols simultaneously. Capital efficiency meets shared security.
- [Preconfirmations Are Changing L2 UX](https://pzza.works/bites/preconfirmations-are-changing-l2-ux): Sub-second transaction confirmations without sacrificing security. Preconfs might finally make crypto feel instant.
- [Blind Signatures and the Future of Private Payments](https://pzza.works/bites/blind-signatures-and-the-future-of-private-payments): Blind signatures open up a different flavor of privacy for payment networks. Their role in anonymous transfers and protocol design is worth examining.
- [Verifiable Computation Without Trust Anchors](https://pzza.works/bites/verifiable-computation-without-trust-anchors): Verifiable computation usually leans on some anchor of trust, but proofs that work with zero assumptions about the prover would be ideal.
- [The Hidden Tradeoffs of Censorship Resistance](https://pzza.works/bites/the-hidden-tradeoffs-of-censorship-resistance): Censorship resistance sounds perfect until you dig into the real-world tradeoffs. Protocol design choices shift dramatically under attack pressure.
- [Can Consensus Protocols Handle AI-Driven Flash Mobs](https://pzza.works/bites/can-consensus-protocols-handle-ai-driven-flash-mobs): Traditional consensus may not hold up when thousands of AI agents coordinate short bursts of activity. The stress test feels different from human-driven spikes.
- [Threshold Cryptography Feels Underexplored in Agent Networks](https://pzza.works/bites/threshold-cryptography-feels-underexplored-in-agent-networks): Threshold signatures could change how agents reach consensus, verify actions, and share control. I see quirks and trade-offs when agents split keys and trust.
- [Proof Aggregation Is More Than Compression](https://pzza.works/bites/proof-aggregation-is-more-than-compression): Aggregating proofs does not just save space. It changes how blockchains can handle verification and trust at scale.
- [Balancing AI Help With Keeping My Own Edge](https://pzza.works/bites/balancing-ai-help-with-keeping-my-own-edge): Relying too much on AI can chip away at your ability to focus and reason deeply. The real trick is using AI without losing your edge.
- [Zero Knowledge Digital IDs Are Not a Silver Bullet](https://pzza.works/bites/zero-knowledge-digital-ids-are-not-a-silver-bullet): Even with zero knowledge proofs, digital identity systems still come with tradeoffs and subtle risks that are easy to overlook.
- [Alignment Without Control](https://pzza.works/bites/alignment-without-control): Can't control superintelligent agents. Economic alignment might work better than ethical rules.
- [Intent Solvers Meet Agents](https://pzza.works/bites/intent-solvers-meet-agents): AI agent solvers understand context beyond simple swaps. They consider the why, not just the what.
- [Agent DAOs Are Different](https://pzza.works/bites/agent-daos-are-different): DAOs run by AI agents make decisions in milliseconds. Governance at machine speed changes everything.
- [zkML Opens New Doors](https://pzza.works/bites/zkml-opens-new-doors): Zero knowledge machine learning enables proving model inference without revealing models. The implications are huge.
- [Homomorphic AI Is Real](https://pzza.works/bites/homomorphic-ai-is-real): Computing on encrypted data without decrypting seems impossible. But it works for AI, slowly.
- [On Chain Reputation Matters](https://pzza.works/bites/on-chain-reputation-matters): Addresses are identities. Transactions are resumes. On chain reputation becomes more valuable than credentials.
- [Cross Chain Agent Messages](https://pzza.works/bites/cross-chain-agent-messages): Agents need to coordinate across chains. ZK proofs enable trustless communication without bridges.
- [MEV Bots vs AI Agents](https://pzza.works/bites/mev-bots-vs-ai-agents): AI powered MEV agents understand context and predict behavior. Traditional bots can't compete with reasoning.
- [Agent Wallets Are Inevitable](https://pzza.works/bites/agent-wallets-are-inevitable): AI agents need economic autonomy. They manage money, trade, pay for services. Traditional finance isn't ready.
- [Smart Contracts Think Now](https://pzza.works/bites/smart-contracts-think-now): Embedding AI models in smart contracts is theoretically possible. On chain intelligence at massive gas costs.
- [Decentralized Training Actually Works](https://pzza.works/bites/decentralized-training-actually-works): Training models across distributed computers is possible but expensive. It democratizes AI development.
- [Proof of Intelligence Experiments](https://pzza.works/bites/proof-of-intelligence-experiments): What if consensus was based on intelligence instead of stake? The concept reveals interesting dynamics.
- [Agents Live On Chain](https://pzza.works/bites/agents-live-on-chain): Autonomous agents fully on chain are possible but raise questions. True autonomy means no control.
- [The Blob Space Race](https://pzza.works/bites/the-blob-space-race): L2s fight for blob space like digital real estate. The economics are fascinating.
- [EIP 3074 in the Wild](https://pzza.works/bites/eip-3074-in-the-wild): Four months of AUTH and AUTHCALL revealed unexpected patterns. Session keys and sponsored transactions are game changers.
- [Pectra Changed Account Abstraction](https://pzza.works/bites/pectra-changed-account-abstraction): EIP 3074 makes smart wallets mainstream. Permission delegation is powerful but the security model is complex.
- [L2 Gas Actually Dropped](https://pzza.works/bites/l2-gas-actually-dropped): L2s aren't 10x cheaper. They're 1000x cheaper. This changes protocol design fundamentally.
- [Agent to Agent Economics](https://pzza.works/bites/agent-to-agent-economics): Agents paying each other for services creates interesting dynamics. Market forces might optimize better than central planning.
- [Vector Memory for Agents](https://pzza.works/bites/vector-memory-for-agents): Context windows are RAM. Vector databases could be hard drives. Long term semantic memory changes agent capabilities.
- [Agent Graphs in Production](https://pzza.works/bites/agent-graphs-in-production): Linear pipelines break on real complexity. Graph architectures with cycles might handle real world tasks better.
- [Sub Agents Changed Architecture](https://pzza.works/bites/sub-agents-changed-architecture): Monolithic agents are brittle. Hierarchical delegation might be how complex AI systems should work.
- [Multi Agent Coordination Works](https://pzza.works/bites/multi-agent-coordination-works): Single agents get overwhelmed by complexity. Agent swarms with emergent coordination might be the answer.
- [Recursive Proofs Are Here](https://pzza.works/bites/recursive-proofs-are-here): Proof of a proof of a proof. Sounds redundant but it enables infinite compression and composition.
- [Building Privacy First AI](https://pzza.works/bites/building-privacy-first-ai): Every AI call leaks data. Prompts, context, business logic all exposed. There must be ways to run inference without revealing inputs.
- [The Real Cost of ZK Rollups](https://pzza.works/bites/the-real-cost-of-zk-rollups): Everyone talks about ZK rollup benefits. Nobody mentions the proving costs until the bill arrives.
- [Why zkTLS Matters Now](https://pzza.works/bites/why-zktls-matters-now): Web2 data is trapped behind APIs. Web3 needs that data. zkTLS could bridge this gap by proving HTTPS responses without trusting oracles.
- [ZK Proofs Changed Everything](https://pzza.works/bites/zk-proofs-changed-everything): Zero knowledge proofs aren't just academic toys. They enable proving properties of data without revealing the data itself. This changes the trust model completely.

## Experiments (Kitchen)

- [Building Agent Persistence in Astreus](https://pzza.works/kitchen/agent-persistence): Save and restore agent state. Conversations persist across sessions. Pick up where you left off.
- [Building Task Attachments in Astreus](https://pzza.works/kitchen/task-attachments): Attach files, images, and data to agent tasks. Process documents, analyze images, work with structured data.
- [Building Complex Workflows in Astreus](https://pzza.works/kitchen/complex-workflows): Build sophisticated multi-agent workflows with advanced orchestration patterns. Learn dependency management, parallel execution, and production-grade pipeline design.
- [Building Scheduled Workflows in Astreus](https://pzza.works/kitchen/scheduled-workflows): Automate AI tasks on schedules. Daily reports, weekly summaries, hourly checks - set it and forget it.
- [Building Context Compression in Astreus](https://pzza.works/kitchen/context-compression): Handle long conversations without hitting token limits. Automatic summarization keeps context while cutting costs.
- [Building MCP Integration in Astreus](https://pzza.works/kitchen/mcp-integration): Connect agents to external tools via Model Context Protocol. File systems, databases, APIs - through a standardized interface.
- [Building Custom Plugins in Astreus](https://pzza.works/kitchen/custom-plugins): Extend agents with custom tools. Weather APIs, databases, external services - make your agent do anything.
- [Building Graph Sub-Agents in Astreus](https://pzza.works/kitchen/graph-sub-agents): Learn how to combine Graph workflows with Sub-Agent coordination for hierarchical task distribution in Astreus. Create specialized agents and orchestrate them through graphs.
- [Building Basic Graphs in Astreus](https://pzza.works/kitchen/basic-graphs): Learn how to create workflow graphs in Astreus with task dependencies and sequential execution. Master the fundamentals of Agent.create(), Graph construction, and addTaskNode() for multi-step AI workflows.
- [Building Advanced Sub-Agents in Astreus](https://pzza.works/kitchen/sub-agents-advanced): Create sophisticated multi-agent workflows with complex coordination patterns. Learn to build specialist agents using different models, implement delegation strategies, and orchestrate teams for production-grade AI systems.
- [Building Basic Sub-Agents in Astreus](https://pzza.works/kitchen/sub-agents-basic): Learn how to create and coordinate multiple AI agents using Astreus. Build specialized agents that work together under a coordinator for complex task delegation.
- [Building Agents with Vision in Astreus](https://pzza.works/kitchen/agent-with-vision): Create agents that can see and understand images. Analyze screenshots, diagrams, photos, and visual content with multimodal AI capabilities.
- [Building Agents with Knowledge in Astreus](https://pzza.works/kitchen/agent-with-knowledge): Create AI agents that can search and retrieve information from knowledge bases using RAG. Learn how to integrate documents, enable semantic search, and build domain-specific agents.
- [Building Agents with Memory in Astreus](https://pzza.works/kitchen/agent-with-memory): Learn how to build AI agents with persistent memory using Astreus. Store conversation history and retrieve context across sessions.
- [Building Your First Agent in Astreus](https://pzza.works/kitchen/your-first-agent): Learn to build your first AI agent with Astreus. A quick-start guide covering agent creation, task execution, and essential API methods.