Your AI Said “Done.” Make It Prove It.
Your AI coding tool says done. story-gate makes it prove each acceptance criterion before merge. A free per-story gate for vibe coding, production ready.
Why agent workflows need a graph, not a queue
The first agent we put into production ran for eleven days before it charged a customer twice. Nothing in the logs looked wrong. The model had done exactly what it was asked, twice, because the process that asked it had restarted halfway through a step.
What a queue cannot tell you
A queue knows that a message was delivered. It does not know that the message was step four of nine, that step three produced a document the model now depends on, or that step four had already called a payment API before the worker died.
We tried to make it know. Every message grew a header with the run id, the step number, the hash of the previous output. Within a month the headers were the system, and the queue was a slow, expensive way to move them around.
What a graph gives you
A graph of typed steps makes three things explicit that a queue leaves implicit.
Order and dependency. Step four cannot start until step three has a recorded result. That sounds obvious; it is not what most agent loops actually enforce.
Idempotency. Each step has a key derived from its inputs. A retry with the same key gets the same result without calling the model or the API again.
An audit trail you can read. When a customer asks why something happened, you walk the graph. Every node has the prompt, the response, the tool call and the timestamp.
const run = graph()
.step('fetch-invoice', fetchInvoice)
.step('match-po', matchPO, { after: ['fetch-invoice'] })
.step('charge', charge, { after: ['match-po'], idempotent: true });What it cost us
Building this took longer than bolting headers onto a queue. It also ended the duplicate charges, and it turned “what did the agent do” from an afternoon of log archaeology into a page.
That is the trade we would make again, and it is the reason GraphSmith exists.
Questions about this article
Does this mean queues are bad?
No. Queues are excellent at moving work between services. They are the wrong place to store the state of a multi-step agent run, which is what we tried to make them do.
Is GraphSmith a replacement for Temporal or Airflow?
No. It sits at the level of a single agent run and can be hosted inside those systems. It cares about steps, their inputs and outputs, and what happens when a step fails.
What does "never duplicates work" mean in practice?
Every step has a deterministic key. If a crash happens after a step completed but before its result was recorded, the retry recognises the key and reuses the result instead of calling the model again.
Sources
Free tool · private betaIs your AI bot doing a good job? Get a free report card.
Ask your bot to check your chat. Agent Tune-Up grades it on 5 simple checks and tells you the one fix to make first. It works inside Grok Bot, Claude, ChatGPT and Cursor. Nothing to install.
How it works: 4 steps
-
Get your free connector
Type your email in the box at the bottom. We email you a link. Open it, and your connector page has everything you need for step 2.
-
Add it to your AI (one time)
Your connector page gives you one message to copy. Paste it into your bot, then paste your private token into the hidden field it shows (never into the chat). Your bot adds Tune-Up itself.
Grok Bot: exactly as above. No settings to find.
Claude Code or Cursor: your connector page has a ready-made line to copy.
Claude.ai or ChatGPT: skip the email. Add a custom connector with the addresshttps://viaknox.com/api/tuneup/mcpand sign in with your email when it asks.
Codex (app, CLI or IDE on your computer): skip the email. Runcodex mcp add agent-tune-up --url https://viaknox.com/api/tuneup/mcpand sign in when your browser opens. -
Ask your bot to check itself
Open your chat with the bot, the way you always do. In Grok Bot, type @ and pick Agent Tune-Up first. Then say:
check our chat so far
Your bot sends Tune-Up the conversation and its own instructions. Nothing to copy or paste.
Chat from another app? Paste it in instead. Your connector page has a fill-in message for that.
-
Read your report card, then paste the fix
In about a minute your bot shows the report card. Each check gets a ✅ or ❌ with the reason, in plain words.
It ends with "What to change" (up to three things, worst first) and the lines to paste, with where to paste them for your app. Paste, save, and check your next chat to see if it worked.
- Free · 25 checks a day
- About a minute per check
- Private: we automatically remove emails, phone numbers and common key and password formats, then send a scrubbed copy to an AI judge; we never keep your chats
- Honest limit: it reads the chat only, so it can't see what your bot did behind the scenes
Try GraphSmith.
GraphSmith is a free, open-source tool that generates AI agent workflows that survive crashes, never duplicate work and don't hallucinate code.
GraphSmith · by Viaknox
GraphSmith is a free, open-source tool that generates AI agent workflows that survive crashes, never duplicate work and don't hallucinate code.