Automate faster and smarter with AI Actions

Coming to everyone on 24th February

---

---

_{Don't want to read? - Watch the video instead:}

---

---

Over the past few months, I’ve been sharing how AI Actions can be used inside Pipelines to

extract insights from emails
analyze complex documents like 10-K filings
and even
generate weekly report summaries automatically

The feedback has been great. You love the flexibility and power.
But it's also been very clear where things could be better.

The release on the 24th will be about answering those questions.

With three major improvements, AI Actions becomes easier to use, more context-aware, and far more versatile — without changing the core idea: one simple step that lets AI reason over your pipeline data directly inside your workflows.

---

What We’ve Done So Far

AI Actions started intentionally simple.

---

Use Case 1:
Smart record matching between emails and Quickbase records

Incoming emails often reference projects by name — but rarely in a clean, consistent way.

For example:

“From this email, find the Quickbase project whose name is mentioned somewhere in the email, and return its project name and record ID.”

With AI Actions, you can:

1. Use the Outlook trigger to capture the email content.
2. Run a Quickbase query to fetch candidate project records.
3. Pass both the email step and the query step to AI Actions.
4. Let the AI match the email text to the correct Quickbase record based on meaning or exact text.

Why it matters:
You avoid brittle string matching and complex branching logic, while still getting a structured result that flows cleanly into the rest of your pipeline.

---

Use Case 2:
Replace complex Jinja with natural language

Most complex pipeline logic is technically possible with Jinja — but painful to build and even worse to maintain.

A classic example: workday date rules, like:

“Given a date, calculate the first working day of the following month.”

Before, this often meant:

• Lots of Jinja
• Nested conditions
• Edge cases (weekends, month boundaries, holidays depending on rules)

With AI Actions, you can describe the rules in plain language and let the AI do the reasoning.

And with the Structured Output Builder, you don’t have to hand-write a JSON schema — you simply define the fields you want (for example: is_working_day, next_working_day, explanation), and the AI fills them in reliably.

Why it matters:
You replace brittle templating logic with a readable prompt + structured output that’s easier to build, easier to change and to understand later.

---

Use Case 3:
Compare two Quickbase tables to find discrepancies

Reconciling data between systems is a common headache — especially when the same list lives in two separate Quickbase tables (or a spreadsheet).

With AI Actions, you can pass two prior steps into one prompt — for example:

• A Quickbase query in app A (your source of truth)
• Another Quickbase query in app B (what a team is tracking)

Then ask something like:

“Compare these two datasets and list discrepancies: missing rows, mismatched values, and any duplicates. Return a clear summary plus a structured list of issues.”

Why it matters:
Instead of exporting files, writing custom logic, or manually scanning rows, AI Actions can do the comparison automatically and return a discrepancy report you can store in Quickbase or send to stakeholders. Even more, you can process it directly in the pipeline.

---

Use Case 4:
Extract invoice details from a photo

Not everything comes in as a clean PDF or spreadsheet — sometimes it’s just a photo taken on a phone.

With image file support in AI Actions, you can now pass common image formats (JPG/PNG/GIF/WEBP) directly to the AI and extract structured invoice details.

For example:

“From this invoice photo, extract vendor name, invoice number, invoice date, total amount, and line items.”

Pair this with the Structured AI Output Builder, and the AI returns the exact fields your pipeline needs — ready to create a record, route for approval, or trigger a payment workflow. How to deal with nested lists/arrays.

Why it matters:
You can turn unstructured image uploads into structured Quickbase data automatically, without manual extraction.

---

Improvements

On 24th February we're launching the latest and biggest batch of improvements to the channel:

• Users on all plans can try AI Actions before April 1, 2026 at no cost
• Significantly increased daily limits for data sent to and received from the AI (per account)
• Increased limits for data processed per step execution
• Increased file size limit to 3 MB
• Structured AI Output simple builder (released 12th Feb)
• Reference entire outputs from prior steps
• Image support for files
• Other performance and usability bugs