BLOG
Writing about what it takes to run AI agents in production
Durable execution, human-in-the-loop, fleet governance, cost control, and coding-agent workflows.
Apr 14, 2026 · code
Agent-Native Durability: Why I Built AXME Code
A few months ago I was caught in a cycle of zero-knowledge cold starts with Claude Code — re-explaining my stack, my architecture, and my safety rules every session. So I built AXME Code: a structured knowledge base, hook-based safety guardrails, and ~10x better token efficiency on LongMemEval than the closest competitor.
Read →Apr 11, 2026 · code
After 100 Sessions: What Persistent Memory Changed About How I Use Claude Code
Before: re-explain the project every session, watch the same mistakes, fix the same bugs twice. After 100 sessions with persistent memory: different patterns, measurable time savings, one problem I did not expect.
Read →Apr 11, 2026 · code
CLAUDE.md Is a Read-Only File. Your Coding Agent Needs Writable Memory.
Everyone puts project context in CLAUDE.md. It's static, it grows, nobody maintains it, and it becomes a graveyard of stale rules. Writable memory with an audit trail is a different shape of solution.
Read →Apr 11, 2026 · code
You've Had This Conversation With Claude Code 47 Times
Every session starts the same way: explaining your stack, your deploy flow, your conventions. You've said it forty times. The agent keeps nodding. Here's why re-explaining every session is the wrong fix.
Read →Apr 11, 2026 · code
Decisions Are Not Memories. Most Agent Memory Systems Confuse Them.
'I learned we use Postgres' and 'we decided Postgres over MySQL because of JSON support' are different data. One is a fact. The other is a contract. Mixing them is why agent memory drifts.
Read →Apr 11, 2026 · code
The First Session With a New Codebase: What AXME Code Does Before the Agent Starts
Stack detection, structure scanning, glossary extraction, safety rule seeding. What happens in the first 30 seconds of a new AXME Code setup, why each step matters, and what happens if you skip it.
Read →Apr 11, 2026 · code
The 4 Types of Context Your AI Coding Agent Needs
Most agents treat all context as 'stuff in the context window'. But facts, patterns, project state, and references have different lifecycles and different retrieval rules. Mixing them is why agents feel dumb.
Read →Apr 11, 2026 · code
Multi-Repo Workspaces Break Claude Code. Here's the Fix.
You have 5 repos in one workspace. Each has its own conventions. A single CLAUDE.md is the wrong shape. Per-repo storage with a workspace root is the pattern that actually scales.
Read →Apr 11, 2026 · code
Your AI Coding Agent Keeps Running `git push --force`. Prompts Don't Fix This.
You put 'never force push' in CLAUDE.md. The agent force-pushed anyway. Here's why prompt-level safety rules fail and what tool-level enforcement actually looks like.
Read →Apr 11, 2026 · code
Stop Writing CLAUDE.md Files. Write Decisions Instead.
CLAUDE.md is a passive artifact. Decisions are active rules with enforcement levels, supersede chains, and audit history. Here's why I stopped maintaining CLAUDE.md and what I switched to.
Read →Apr 11, 2026 · code
Subagents Without Context: Claude Code's Silent Bug
When Claude Code spawns a subagent to run a task in parallel, the subagent starts fresh. It doesn't inherit the parent's memory. Work gets done, but context is lost. Here's the fix, and why it's fragile.
Read →Apr 11, 2026 · code
I Went on Vacation. My Agent Came Back to an Empty Project.
Five days off. Session handoff gone. Half a week of decisions forgotten. Here's the 'what was I doing yesterday' problem and why session close is a ritual that matters.
Read →Apr 11, 2026 · code
We Built AXME Code With AXME Code. Here's What the Memory File Looks Like.
Six months of dogfooding. Real memories, real decisions, real safety rules from our own repo. What patterns emerged. What we threw away. What surprised us.
Read →Apr 8, 2026 · cloud
Your AI Agent Crashed at Step 47. Why Isn't Crash Recovery the Default?
Your agent ran 47 steps of a 50-step pipeline. Then it crashed. The state is gone. Every framework says you should have configured checkpointing. Why isn't durability built in?
Read →Apr 8, 2026 · code
How I Fixed Claude Code's Amnesia (Actually)
Claude Code forgets your project every session. CLAUDE.md helps, but it doesn't scale. I built an MCP server plugin that gives Claude Code persistent memory, decisions, and safety guardrails.
Read →Apr 8, 2026 · cloud
How to Add Human Approval to AI Agent Workflows Without Building It Yourself
Adding a human approval step to an AI agent workflow means building a notification service, reminder scheduler, escalation chain, and webhook handler. Or you can use 4 lines of code.
Read →Apr 8, 2026 · mesh
How to Stop a Rogue AI Agent in Production
Your AI agent went rogue at 3am. It's running on multiple instances across regions. There's no terminal to Ctrl+C. You need a kill switch that works in under 1 second, enforced at the infrastructure level.
Read →Apr 8, 2026 · cloud
Temporal Alternative Without the Cluster and Determinism Constraints
Temporal is the go-to for durable execution. But it requires a cluster and forces determinism constraints on your workflow code. Here's an alternative that gives you durability through a managed API with none of those trade-offs.
Read →Apr 5, 2026 · mesh
You Have 50 AI Agents Running. Can You Name Them All?
AI agents are multiplying across your org. Different clouds, different frameworks, different teams. Without a fleet dashboard, you're flying blind.
Read →Apr 2, 2026 · mesh
Your AI Agent Spent $500 Overnight and Nobody Noticed
AI agents call LLMs. LLMs cost money per token. Nobody tracks it per agent. One runaway loop and your OpenAI bill is a disaster.
Read →Apr 2, 2026 · mesh
You Deployed 30 AI Agents. Can You Answer These 5 Questions About Them?
Most teams can't tell you which agents are running, how much they've spent, or how to shut one down. Here's what a governance platform for AI agents looks like.
Read →Apr 2, 2026 · mesh
3 of Your AI Agents Crashed and You Found Out From Customers
Your agents are running across 4 machines. One dies. No alert. No log. You find out 3 hours later from a customer complaint. Here's how to fix that.
Read →Apr 2, 2026 · mesh
Your AI Agent Stopped Responding 2 Hours Ago. Nobody Noticed.
Container is green. Process is running. But your agent stopped processing work 2 hours ago. Heartbeat monitoring catches what health checks miss.
Read →Apr 2, 2026 · mesh
Your AI Agent Is Running Wild and You Can't Stop It
AI agents go rogue. They send thousands of emails, make unauthorized API calls, burn through budgets. You need a kill switch that works in under 1 second.
Read →Apr 2, 2026 · mesh
Your AI Agent Did Something It Wasn't Supposed To. Now What?
AI agents ignore restrictions in prompts. Gateway-enforced action policies block unauthorized actions before they reach the agent - no prompt injection can bypass it.
Read →Apr 2, 2026 · mesh
Your AI Agent Made 10,000 API Calls in an Hour. Here's How to Stop That.
One runaway retry loop. 10,000 API calls. $130 in LLM costs. No rate limit fired because you never built one. Here's how to add centralized rate and cost limiting to AI agents.
Read →Mar 29, 2026 · cloud
A2A Tells Agents How to Talk. It Doesn't Tell Them What Happens When Things Break.
Google's A2A protocol handles agent communication. But crash recovery, retries, timeouts, and human approval gates? That's still on you. Unless you add a lifecycle layer.
Read →Mar 29, 2026 · cloud
Your Agent Has Been Stuck for 3 Hours and Nobody Knows
API call hangs. Agent blocks. No timeout fires. No one gets paged. The default behavior in most agent frameworks is silence. Here's how to fix that.
Read →Mar 29, 2026 · cloud
A Two-Step Approval Chain Shouldn't Need a Workflow Engine
Manager approves, then finance approves. Simple enough to describe. 300 lines of code to build. Unless your platform handles approval chains natively.
Read →Mar 29, 2026 · cloud
Your AutoGen Agents Can't Talk Across Machines. Here's the Missing Piece.
AutoGen handles multi-agent conversations beautifully - inside one process. Put agents on different machines and you're back to building message brokers from scratch.
Read →Mar 29, 2026 · cloud
I Stopped Building Webhook Retry Logic. Here's What I Use Instead.
Exponential backoff, jitter, dead letter queues, idempotency keys, HMAC verification - all to deliver one message reliably. There are better options now.
Read →Mar 26, 2026 · cloud
Your AI Agent Crashed at Step 47. Now What?
AI agents do multi-step work. They crash. The state is gone. Here's how to make agent operations durable without writing checkpoint code.
Read →Mar 26, 2026 · cloud
Why Your AI Agent Shouldn't Block When It Needs Human Approval
AI agents get stuck waiting for humans. There's a better pattern than blocking - async approval with reminders, escalation, and timeout.
Read →Mar 26, 2026 · cloud
What Temporal Can't Do: Human Approval Mid-Workflow
Temporal gives you durable execution. But adding a human approval step mid-workflow requires building a signal handler, notification service, and UI. There's a simpler way.
Read →
GUIDES
Long-form references for durable execution, HITL, governance, and multi-agent coordination.
Ship your first durable agent — in under 10 minutes.
Free tier. No credit card. Self-host or hosted — your choice.