24 to 30 October 2026
This week we shipped Autopilot-from-a-step so recordings can start mid-process, fixed email triggers that were firing thousands of times on the wrong messages, gave every pool its own built-in queue, made connector pages show whether each connection is actually working, and improved how runs are launched so typed inputs are no longer silently dropped.
Autopilot can now record from any step in an existing process
When a process already does the first half of the work, recording the whole thing from scratch wastes time and risks breaking what is already there.
You can now right-click any step on your process map and choose "Autopilot from here". The bot runs the process up to that point for real, keeps the screen open exactly as it would be mid-run, and Autopilot records whatever comes next into the draft. Nothing is promoted automatically. A person reviews the new steps before they go anywhere near a live run.
If Autopilot gets stuck on a step it cannot figure out, it asks for help on the page instead of failing. You can pick the element on a copy of the screen, tell it to click, type or choose, and the bot carries on and saves that as a step.
This means you can extend a working process without re-recording the parts that already work, and without leaving the bot to fail silently when a screen is unfamiliar.
#Automize #DigitalWorkers #ProcessAutomation #Autopilot
Email triggers now check importance and read state before firing a run
One of our customers had a process that ran over 2,400 times in three days, every 30 seconds, with nothing to do each time. Its trigger fired on every unread email. The first step inside the process only looked at emails marked Normal importance, so urgent emails were never handled, stayed unread, and kept firing the trigger.
The trigger and the first email step now warn you when they disagree, before a run ever starts. The "For each email" loop also has new filter fields: read state, importance, subject, sender and whether the email has attachments. A dropdown beside each filter row lists the available fields with plain labels, and for fields with fixed values, such as importance, a second dropdown offers the options directly.
On the bot side, every Outlook step now reads its result back and fails the step if the write did not actually happen, so a move, a flag or a category change that Outlook quietly ignored no longer passes as a success.
#Automize #Outlook #EmailAutomation #ProcessAutomation
Every pool has its own queue from the start, with a deadline option
When every machine in a pool was busy, waiting runs were held in a queue that appeared without warning, had its own separate settings, and showed up in the queues list as an unexplained entry with the pool's name.
Every pool now has one queue built in from the moment the pool is created, the way the leading automation platforms work. Waiting work is always kept, never refused. The queue section lives on the pool's own page, and the pool's retry and backlog alert settings are the queue's settings, so there is one place to look.
A new option lets you set a deadline on waiting items. If a run is still waiting when its deadline passes, it is cancelled and the right person is told. Items already being worked are left alone.
Pools shared with other companies keep their own separate queue, so billing and data stay separate. The queues list no longer shows pools' built-in queues, which keeps it tidy.
#Automize #ProcessAutomation #BotPools #Automation
Connector pages now show whether each connection is actually working
It was easy to set up a connection to another system and not know whether it was still working until a run failed.
Every connection in your connectors list now shows a status on the right: Working, Not working, Not set up, or Not checked. For Anthropic and OpenAI it says which connection is the one in use as the AI key. A connection's own page has a Status box showing the last error and which processes use it. If a connector tile has a connection that needs attention, it says so.
The status comes from the last time Automize actually used each connection, whether that was a cloud step, a test, a sign-in renewal or a live sample. Only a refused sign-in, an unreachable system or a server error counts as not working. A "not found" or a bad input still proves the connection signs in fine.
We also fixed a bug where the Test button on Google Drive, Teams, Outlook and SMTP connections always said a field was missing, because their real fields were not being checked properly.
#Automize #Connectors #Automation #DigitalWorkers
Typed run inputs and variable names with spaces now reach the bot reliably
Two separate problems were causing run inputs to be silently dropped, and both are fixed this week.
The first: running a process from the bot page or pool page built the inputs in a format that a form submission would handle, but a structured request body does not expand the same way. The server saw an empty inputs map and refused the run, even though the person had just typed a value. The fix sends inputs as a proper nested object, which is what the server expects.
The second: variable names with spaces, such as "Company Code", were being dropped by the bot with no log line. Every sub-process that depended on the value compared against a blank and reported success having done nothing. The bot now accepts names with spaces, and the web adds a process check that warns when a variable name has a space or a character the bot reads as a path, with a one-click rename that fixes it everywhere in the process at once.
#Automize #ProcessAutomation #Automation #DigitalWorkers