Product updates, playbooks and stories from teams building with Atwork.
A practical rule for deciding between a workflow automation and an AI agent, plus a comparison table covering cost, reliability and how each one fails.
Step-by-step: scope, persona, statuses and review loop for an AI agent that triages incoming support requests and hands each one back to a human.
A practical way to pick the model behind your AI agent: match the tier to the job shape, start cheap, and upgrade only when you can point at a failure.
Build one real automation end to end: pick the trigger, add a condition so it does not fire on everything, choose the action, and test it without spamming your team.
Table, board, calendar, timeline or gantt — three diagnostic questions that tell you which view fits your project, and the signs you picked the wrong one.
Why sheets decay, and the field-design rules that prevent it: one field per fact, when to use a select, when to split into a second sheet, and how to clean up sprawl.
A minute-by-minute plan for evaluating a work platform: bring in real work, fix the view, invite one colleague, automate one habit, and know by the end whether it fits.
How spaces, sheets, items, fields and views fit together, what decision belongs at each level, and which choices are hard to reverse once your team is using them.
One worked example — clients and projects — carried through all four connected field types, so you can tell which one your problem actually calls for.
Which choice-shaped field to use and what breaks if you pick wrong — why Status is special, the one-vs-many test, and when a checkbox is a mistake.
How to choose between Employee, Team, Job, their list variants, Contact and Attendees — and why this is the most expensive field decision to reverse.
Running an MCP server in production is mostly authorization and tool design, not protocol work. Lessons from putting a real workspace behind MCP.