惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

aimingoo的专栏
aimingoo的专栏
WordPress大学
WordPress大学
阮一峰的网络日志
阮一峰的网络日志
博客园 - 司徒正美
月光博客
月光博客
宝玉的分享
宝玉的分享
Recent Announcements
Recent Announcements
小众软件
小众软件
H
Hackread – Cybersecurity News, Data Breaches, AI and More
美团技术团队
博客园 - 三生石上(FineUI控件)
A
About on SuperTechFans
J
Java Code Geeks
云风的 BLOG
云风的 BLOG
罗磊的独立博客
大猫的无限游戏
大猫的无限游戏
IT之家
IT之家
Vercel News
Vercel News
量子位
Martin Fowler
Martin Fowler
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
V
Visual Studio Blog
腾讯CDC
有赞技术团队
有赞技术团队

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant
Your AI Agent Ordered Bananas. Here's Why.
Kriday Dave · 2026-05-15 · via DEV Community

Kriday Dave

Your AI Agent Ordered Bananas. Here's Why.
You told your AI agent you like apples and oranges. Somewhere between Agent 1 and Agent 2, a banana snuck in. Now your grocery agent is ordering bananas. Not because the LLM hallucinated -- because the context handoff was broken.
This is the real problem nobody talks about. Everyone's arguing about which model is smarter while the plumbing underneath is leaking.
We blame the LLM for everything
Yelling at Claude at 3AM to fix your broken pipeline isn't going to work. Claude doesn't know you. Claude doesn't remember what Agent 1 said to Agent 2 three steps ago. Each handoff between agents is just a JSON blob getting passed around with zero guarantees. No signing. No validation. No rollback if something goes wrong.
RAG gets thrown around as the solution to everything but RAG just helps agents find relevant information. It doesn't protect the information that's already been passed around and corrupted.
If your AI agent corrupts your context -- your Obsidian vault, your memory store, whatever you're using -- every downstream agent inherits the damage. Silently.
The banana problem, formally
Here's what actually happens without any protection:
python# Agent 1 produces output
agent1_output = {"entities": ["Apple", "2024 revenue"], "summary": "Apple grew"}

Manual serialization -- easy to lose data, corrupt context

context = json.dumps(agent1_output)

Agent 2 receives corrupted context

agent2_input = f"Given: {context}\nAnalyze this."
Agent 2 might drop keys. Add hallucinated ones. Quietly rewrite history. You won't know until something downstream breaks in a way that's nearly impossible to trace back.
What if we treated context handoffs like database transactions?
That's the question that led to Relay.
Relay treats your context like an append-only ledger. Every handoff between agents gets signed and validated. If an agent drops a critical key -- like "entities" disappearing between steps -- Relay catches it and rolls back to the last clean state automatically.
pythonfrom relay.core_pipeline import CoreRelayPipeline

pipeline = CoreRelayPipeline(
signing_secret="your-secret-key",
token_budget=8000
)

Agent 1 creates a signed envelope

result = pipeline.execute_step({"entities": ["Apple"], "revenue": "2024"})
envelope1 = result.value # signed, immutable

Agent 2 accidentally drops "entities"

result = pipeline.execute_step({"summary": "growth"}) # contradiction detected

Relay rolls back automatically

result = pipeline.rollback()
restored_envelope = result.value

Back to the clean state from step 1

No banana.
What Relay actually does
Every context move between agents gets wrapped in a signed envelope with a version, a timestamp, the token budget, and an HMAC signature. If anything changes, the signature fails. If a critical key disappears, the validator fires. If the validator fires, you get a rollback to the last clean checkpoint instead of silent corruption propagating through your pipeline.
It also supports parallel agents through fork-join -- you can run multiple agents against the same context snapshot at the same time, then merge or vote on their outputs. If there's a conflict, the whole step rolls back cleanly.
Why this matters
Reliable AI systems need reliable infrastructure underneath them. The model quality is only as good as the context it receives. Right now most multi-agent pipelines have no protection on context handoffs at all.
I built Relay as a 10th grader because I kept running into this problem and couldn't find anything that solved it properly. Context matters. It needs to be protected, not just passed around and hoped for the best.
You can use it today:
pip install relay-middleware
Or check the source: https://github.com/kridaydave/Relay