Make’s January 21 webinar description emphasizes combining agents with structured automation and governance. The recording is older than the current news cycle. This draft has not verified its full demonstrations, so it does not adopt the description’s broad predictions or numerical performance claims.
One judgment, several fixed rules
Imagine a fictional wholesaler receiving supplier invoices by email. AI might suggest whether a free-text message concerns an invoice, a delivery problem or a general question. The decision to approve payment needs a different control: supplier identity, invoice number, purchase order and an authorized approver.
Separate classification from authority. A model-generated category should not create a new supplier or authorize a payment. An unknown category belongs in a review queue rather than a default path that takes an irreversible action.
Write the deterministic contract
For this proposed workflow, require the source message reference, document identifier, extracted fields and validation outcome. Check duplicates before creating another task. Keep the original attachment accessible to the reviewer. If required information is missing, name the exception and owner.
The workflow should record which step completed. If the notification fails after the review task exists, recovery should not create the same task again. Inspect the destination before retrying an uncertain write. These safeguards require configuration; automation does not guarantee them by default.
Evaluate the judgment separately
Use approved samples with clear and ambiguous categories. Compare the suggested classifications to a human-reviewed reference. Track the unknown cases and incorrect routing, then decide whether a simple form field or keyword rule would be easier to maintain.
For a small, stable list of suppliers and document types, ordinary rules may be sufficient. AI becomes interesting where variable language creates meaningful review work. Its usefulness depends on the whole process, including the cost of checking mistakes.
Current product checks
The January 21 recording predates the newer Make AI Agent (New) app. Current Make documentation identifies that app as an open beta released February 2, 2026. It is available on all plans using Make’s AI Provider; custom AI-provider connections require a paid plan. Availability does not mean unlimited free AI use. Agent credit accounting depends on the provider and activity: using Make’s provider includes token-based consumption, while a custom provider can incur separate model charges alongside Make credits. Confirm allowances and the enabled app in the intended account. This entry quotes no historical price as a current offer and does not claim a tested production implementation.
Our Make profile provides the broader platform context. The proposed invoice example is a design exercise, not an accounting system or payment authorization procedure.
Review scope and credit
Video credit: Make. Original title: [Webinar] Top AI and automation trends for 2026. YouTube’s expanded watch page confirms publication on 2026-01-21. Source attribution, description and draft-preview playback were checked November 1, 2026. Automated audiovisual scene analysis was used; this was not a complete manual viewing, and it does not establish that every demonstration or claim was independently verified. Current official documentation was checked for the product claims discussed here. The fictional business scenario and acceptance criteria are our editorial analysis, not a customer result or hands-on product test. Vendor-promotional outcomes and performance figures are not adopted as measured results by this publication.
Watch the original video from Make · Official product information