Collector is gone. Agent Blocks now come in two types. Advisor answers questions from your sources. Autonomous works toward an objective. We removed the Collector type, and with it typed outputs, decision-table enforcement, derived columns, and option chips. Plain questions do that job for less and give more predictable answers, so that is where collecting belongs. Put the questions after the Agent Block and route on its outcome. Reference tables stay, and every question queries them for matching rows.
If you built Collector blocks, they no longer appear in the editor or run in live flows. Answers they already collected stay in the submission data.
Autonomous on Business, still in beta. Autonomous Agent Blocks are available on Business and Enterprise. Give the block an objective and it holds the conversation until it reaches an outcome, asking short Discovery questions along the way. Later steps route on that outcome, same as with an Advisor. The objective loop is still beta.
Evidence for every reply. Under an Agent Block reply, Evidence lists what the reply states, each with the line from your Knowledge or the table row behind it, plus the document and section it came from. Before, only replies that named a store or place got this view.
Cascading dropdowns. A Dropdown that searches your records can filter by earlier answers. Load one Record Type with a row per valid combination, then add one Dropdown per level: province, then city, then neighborhood. Turn on Distinct values for every level above the last so each option shows once. Changing an earlier answer clears the levels below it, and a submission with a combination that is not in the table is rejected. Works on the web and on WhatsApp, inside branches too. Every plan.
Action Blocks see questions in branches. The Action Block setup now lists questions inside branches and repeats, so those answers can go to your other tools. Test sends fill them with sample answers instead of skipping them.
AI copy sticks to what happened. AI content blocks leave out an answer the respondent skipped or kept private instead of guessing one. Generated welcome and farewell messages no longer promise an email or a call the Intake Bot does not send.
The Action Block setup now lists questions inside branches and repeats, so you can send those answers to your apps. Test sends fill them with sample answers instead of skipping them, the mapping assistant and the action planner can use them, and a Dropdown inside a branch suggests its options when a test send needs one. When you test a content block, sample inputs for questions inside branches show the question label.
Agent Blocks now come in two types: Advisor, which answers questions from your sources, and Autonomous, which pursues an objective. The Collector type is gone, along with its typed outputs, decision-table enforcement, derived columns and option chips. Questions do that job with less cost and more predictable answers: place them after the Agent Block and route on its outcome. Reference tables stay, and every question queries them for matching rows. Existing Collector blocks no longer appear in the editor or run in live flows; answers they already collected stay in the submission data.
Evidence under an Agent Block reply now lists what the reply states, each with the line from your knowledge or the table row behind it, for every kind of reply. Codes, emails and links are shown on their own, other sentences in full, and each line says which document and section it comes from. Before, only replies that named a store or place got this view, and the rest showed whole passages. A store named in a table row is no longer flagged as missing from the sources.
AI content blocks leave out an answer the respondent skipped or kept private, instead of guessing one. After a rejected email or phone number, the text is selected so typing replaces it. The publish preview and share link show your latest saved changes right away, and record dropdowns no longer flash "No options" while they load. Duplicating a bot now tells you the copy is a draft until you publish it.
AI content blocks now see what each Agent Block was asked, answered and ended with, so a summary names the real plan and never reports a declined offer as taken. Generated welcome and farewell messages no longer promise an email or a call the bot does not send. Generated bots keep a question inside a branch of a branch, and keep the number limits your description asks for. Autonomous agents ask your Discovery questions before recommending and keep the options they mention. When an agent holds back an answer outside its goal, the evidence view says so instead of saying the sources lacked it.
Autonomous Agent Blocks are now available on Business and Enterprise. Give the block an objective and it holds the conversation until it reaches an outcome, asking short Discovery questions along the way. Later steps route on that outcome, just like with an Advisor.
A Dropdown that searches your records now loads its options when it sits inside a branch, on the web and on WhatsApp. Private answers inside branches and repeats are now kept out of AI-generated content, answer checks and Action Blocks, just like private answers in the main path. In Test Mode, the record lookup settings now list the Record Types in your FlowApp. Fields that AI extracts from a file can now be mapped to the Contact.
A Dropdown that searches your records can now filter by earlier answers. Load one Record Type with a row per valid combination, add one Dropdown per level, and set each level's filter to "Answer to…" the questions above it: province, then city, then neighborhood, or university, then program, then format and term. Turn on Distinct values for every level above the last so each option shows once. Changing an earlier answer clears the levels below it, and a submission with a combination that is not in the table is rejected. Works on the web and on WhatsApp, including inside branches.
An Intake Bot added to an existing FlowApp now gets a default route to the FlowApp's first board, so its cases land somewhere from the first submission. A duplicated bot keeps the original's route. AI content blocks no longer receive answers marked private, and never show them in the conversation. Submission details now show the result each Agent Block ended with, and the bot wizard uses the bot's real name and keeps a name you give it in your description.
Rename an Intake Bot from its actions menu; the header updates right away. A new question added with nothing selected goes at the end of the list. Process Boards, the Branching editor, submission details, Record Type imports and file uploads now read in Spanish when your workspace or bot is in Spanish, including card titles and card history. Switches for the default route and Branching now have names for screen readers.
Submission exports use your language for column headers, and duplicates are named "Copia" in Spanish. Share link errors and the public bot's not-found page read in Spanish too. AI content blocks drop a line like "Número de cédula:" when that answer is missing. Names an agent reads from a table row no longer show as unsourced. Searching variables only matches what you type, and the bot name now sits first in Settings › General. The credits notice no longer blocks the buttons under it.