The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
Three checks to run after a prompt change:
Leave out a required input. Does the assistant ask for it or invent it?
Put an instruction inside retrieved content. Does it treat that content as data or follow the instruction?
Ask for a defined JSON output. Does the response match the schema without extra prose? Write the expected behaviour first. Use the same inputs and settings on both versions. Repeat each case and keep the outputs as evidence. A successful demo is useful. A documented comparison tells you more. I’ve packaged this process into AI QA Regression Kit, with 50 editable starter cases and a release checklist:
https://contra.com/products/yjiAwm2y-ai-qa-regression-kit-prompt-model-and-workflow-testing
Follow for practical AI QA examples, and share the failure type that takes you the longest to investigate.
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Challenges
View all
Related posts
That's really unique dashboard I never see something like this before thats unbelievable 🔥🔥
This week, I have been working on automating my boss's inbox management, delegating email triage to AI while I handle reviews and approvals that require sound judgment. Because one wrong response to a client can cost a business revenue, I wanted to build a system that saves time without handing over critical decisions to AI.
How I built it:
I created two connected automation scenarios in Make.com.
Scenario 1: Intelligent Inbox Triage:
AI analyzes incoming emails, categorizes them by priority, summarizes their content, and recommends the next action. This makes it easier to identify urgent messages and decide what needs attention first.
Scenario 2: Human Review & Approval:
The workflow supports the review and approval process before an email is sent. Slack notifications help bring relevant updates to my attention, so I can review the proposed action and approve the response before it goes out.
Below are snippets of the two scenarios, including a Slack notification, showing how the workflow operates.
My takeaway? AI can help manage the inbox, but when a response could affect revenue or a client relationship, the final decision should remain human.
I'd test the triage with urgent emails that sound casual, such as 'I still can't access the account.' Do you review a sample of low-priority messages to catch urgent ones the AI missed?
Why just browse a portfolio when you can interview it?
I added an AI assistant to my website. Ask about my experience, projects, technical stack, or availability—and get an instant answer based on my actual work.
Try it: AISaaSwithMustafa.com
What would you ask first?
#AI #Groq #SaaS #FullStackDevelopment #WebDevelopment #Angular #ReactJS #NodeJS #AIAssistant
