Instruct

Instruct's pitch is the absence of a builder. Instead of mapping triggers to actions, you write what you want done and the agent executes across connected apps, pulling context from one system to act in another. The clearest example on the vendor's own site is the operations play: read every invoice sent to your email this month, extract vendor, date and amount into the Expenses Google Sheet, and upload the PDF to Drive. That is three apps in one request, and it is the kind of job most non-technical operators currently do by hand or hand to a developer. The second half of the product is what most one-shot AI agents lack. Any successful task can become a permanent process simply by asking Instruct to run it on a schedule. That converts a one-off request into a recurring workflow without you designing anything, and it is the reason the tool belongs in a recurring-operations conversation rather than a chat-assistant one. Where it fits: operations teams pulling invoices, expenses and files out of email into Sheets and Drive; sales reps monitoring HubSpot for leads quiet for three days, researching company news and saving a personalized draft; recruiters filtering the last 50 Ashby applicants for 4+ years of Python and adding them to a Phone Screen shortlist; educators reading Drive lecture notes, generating a 10-question revision quiz and emailing it out. All four are documented on the vendor site with the exact prompt. The honest weaknesses. Coverage is narrow and named only by example — Gmail, Google Drive, Google Sheets, HubSpot, Salesforce, Ashby. If your process touches a ticketing system, a data warehouse, or an internal tool, the task cannot run end to end, and agentic execution degrades badly when one hop is missing. Second, the pages document no audit trail, no guaranteed step order, and no error handling; for finance and compliance workflows where you must show what happened and in what sequence, that is a real gap, and deterministic tools remain the safer choice. Third, the underlying model is not named anywhere on the scraped pages, so you cannot reason about capability or data handling from the documentation. Fourth, no offline mode, desktop client, self-hosted deployment, compliance tier, or separate API documentation appears on the scraped pages, so developer-oriented and regulated buyers should look elsewhere. Positioning: it competes with Zapier and the newer AI agent platforms, replacing trigger-and-action configuration with conversation. The generator side — documents, slides, spreadsheets, images, audio, video, podcast episodes — makes it useful even when the cross-app automation is a stretch. Start with one recurring chore you know you can verify by eye, run it for a week, and only then hand it something that touches money.