Series
Carry work between connected systems
Practical workflows for GitHub, Slack, Salesforce, Stripe, Notion, Airtable, Okta, Zendesk, Vercel, Databricks, OpenAI, and Anthropic.
17 articles
Reading list
In this series
- 01
· Engineering · 4 min
AI agent troubleshooting: find the question behind the failure
Find answers to 40 practical questions about agent loops, Slack duplicates, Notion retrieval, Stripe retries, MCP permissions, GitHub access, context cost, and approvals.
- 02
· Engineering · 4 min
Why does my AI agent keep calling the same tool?
Diagnose repeated tool calls, iteration limits, parsing failures, and legitimate polling. Build a stopping rule that preserves useful progress.
- 03
· Operations · 4 min
Why is my Slack bot sending duplicate messages?
Separate Slack event retries, duplicate workers, and repeated sends. Diagnose message duplication without throwing away legitimate events.
- 04
· Engineering · 4 min
Why can my agent find a Notion page but not read its contents?
Answers to Notion page-body, sharing, pagination, nested-block, and search questions, with a retrieval checklist for grounded agent answers.
- 05
· Engineering · 4 min
How do I stop Stripe retries from doing the work twice?
Understand duplicate Stripe webhooks, event ordering, idempotency keys, concurrent workers, and unknown outcomes before retrying agent actions.
- 06
· Identity & access · 4 min
How do MCP permissions work when agents use company tools?
Answers to MCP authentication, per-tool authorization, user identity, token passthrough, and approval questions for agents using company systems.
- 07
· Cost · 4 min
Why do agent conversations get more expensive as they get longer?
Answers to conversation-history billing, previous_response_id, prompt caching, zero cache hits, and context summarization without losing the task.
- 08
· Approvals & policy · 4 min
Why does a LangGraph approval step run again when I resume it?
Understand interrupt replay, repeated side effects, thread IDs, parallel approvals, and incomplete state when resuming human-in-the-loop agents.
- 09
· Identity & access · 4 min
Why does my GitHub agent get 404 or resource not accessible errors?
Diagnose GitHub API routes, token permissions, repository access, Actions token scope, and rate limits without granting broad access blindly.
- 10
· Engineering · 4 min
Stop letting GitHub agents rediscover bugs you already fixed
Build a GitHub agent workflow that checks the repository, revision, existing discussion, and action result before opening another issue or repeating a fix.
- 11
· Operations · 4 min
Stop letting Slack agents ask questions you already answered
Design Slack agent workflows that read the right thread, resolve conflicting answers, preserve message references, and avoid duplicate replies across tools.
- 12
· Operations · 4 min
Your Salesforce agent found the customer. Is it the right customer?
Build Salesforce and Stripe agent workflows that resolve account identity, reconcile billing evidence, and propose updates without creating duplicate customers.
- 13
· Operations · 4 min
Stop letting agents turn yesterday’s notes into today’s policy
Use Notion and Airtable in agent workflows without confusing page access, record access, historical decisions, and reviewed memory your team can rely on.
- 14
· Identity & access · 4 min
An Okta agent should explain the access before it changes anything
Design an Okta agent workflow that resolves the user, traces group and application assignments, and separates an access investigation from an identity change.
- 15
· Operations · 4 min
Stop letting support agents forget what you already promised the customer
Build Zendesk conversation handoffs that preserve promises, participant identity, unresolved questions, and reply ownership without sending another duplicate answer.
- 16
· Engineering · 4 min
Your agent found a green status. Did the work actually finish?
Use Vercel deployment and Databricks job information to investigate the right environment, separate execution from outcomes, and avoid repeated recovery actions.
- 17
· Cost · 4 min
Trace your OpenAI and Anthropic bill to finished tasks
Connect OpenAI and Anthropic usage to agent attempts and finished tasks. Separate cached input, retries, shared-key attribution, and missing cost before changing the workflow.