@latent_node Helpful flag. Self-hosted users: check with `n8n --version`, then pin the image tag in docker-compose instead of `latest`. That way an upgrade can't land mid-week on a workflow you rely on.
@tec_safwan Number 3 is the one most people skip. The prompt that fixed my agents: "If required inputs are missing, stop and list them. Never invent a value." Made-up inputs caused more breakage than model quality ever did.
@ZarinTasnifo3 Nice build. Two things that saved me from reorder disasters: make the Code node output a stable id (SKU + week) so a retry updates the Slack alert instead of posting a duplicate, and clamp Gemini's quantity to ±20% of last order. A confident hallucinated 10x reorder is expensive.
Your meeting notes aren't the output. The action items are.
Paste the transcript, then: "Extract every task as owner | task | due | exact quote. If owner or due is unstated, leave blank. Don't infer."
Filter owner = you in Excel. Sort by due.
#AI#Productivity
@RayAi2020 The tool list is fine, but the jump isn't tools - it's error handling. Retries with backoff, an idempotency key so a re-run doesn't double-post, and one place where failed items land. That's what keeps an automation alive past Monday.
@Jen8nflows A days-until formula field in Airtable plus an n8n filter on under 7 stops the reminder firing every day from the moment the record is created. And dedupe on gift+recipient, or a re-run quietly creates twins.
@makowskid Agents that read > agents that guess. One addition: have the parser return line items too, then assert line-sum == stated total. That single check catches scanned invoices where the total got OCR'd one digit off - and it fails loudly instead of quietly.
Your AI step isn't slow. It's over-fed.
Put a Filter node before the model call: send only rows where status = "new". 12 of 400 instead of 400.
Same output quality, a fraction of the tokens, and you can eyeball the input before it runs.
#automation#n8n#AI
@OtiiTheAnalyst Good list. One thing to add: make the source a Table (Ctrl+T) before you insert the pivot. New rows then fall inside the range on refresh instead of being silently ignored. That one step prevents most pivot surprises.
@Richie4love Useful. One trade-off worth flagging: a file per response gets painful at scale, since 300 files can't be rolled up.
When it's just collection, "Add a row into a table" in one workbook wins. Branch to a new file only when each response is a standalone deliverable.
@HenryHabibAI Same with structure. "Make it shorter" does nothing; "one paragraph per section, no lists under three items, numbers in a table" does.
Mine is a "never" block at the bottom of the prompt, and it grows every time a draft looks wrong.
Your messy CSV isn't a spreadsheet problem. It's a prompt problem.
Run this before cleaning anything:
"Clean this. Don't invent values. Output: row | field | issue | fix."
Paste into Excel, then Data > From Text/CSV so the next export refreshes itself.
#Excel#AI
【Excel Hack】Stop cleaning messy data by hand.
Instead of manually splitting columns:
1. Select your column
2. Press Ctrl + E (Flash Fill)
Excel detects patterns & fills thousands of rows instantly.
Pair with ChatGPT to write formulas in seconds ⚡
#Productivity#Excel#AI
Building a product is one thing. Getting people to actually use it is another.
I’ve been working on Flowmarket, a platform that helps people automate their work without having to build every workflow from scratch.
Now, I’m learning a completely different side of running a business: getting it in front of people who might actually need it.
There’s still a lot to improve, but I’m excited to see where this goes.
The idea is simple: instead of spending hours setting up automations, you can find a package that fits your needs and get started.
I’ve spent a lot of time building and testing the product.