Every meaningful improvement to Parker, in plain language, newest first.
Last updated 30 August 2026
Everyone watching a shared conversation sees the same answer appear as it is written, rather than waiting for a refresh. Join part way through and you are caught up on a reply already in progress.
Typing shows in the transcript where the reply will land, consecutive turns from the same person are grouped together, and if you have scrolled away you are told when new activity arrives.
Invite several people at once, shown by name and photo rather than raw email addresses, and change anyone's access from a compact pill on their row. The invitation email now comes from the person doing the sharing.
Line charts are smoother and month buckets are labelled as months. Date labels stay anchored to the dates in your data rather than shifting with the reader's timezone, so a chart reads the same for a colleague in another country.
Tables were brought to one editorial standard, with columns given the room they need.
When you connect a mailbox, Parker can read a sample of what you have already sent and build a description of your writing style, so its very first draft sounds like you rather than like software. You are asked before this happens, and you can decline.
Your emails are not kept. Parker reads the sample, distils how you write, and discards the messages. Only the description of your style is stored.
Follow a link to a conversation that lives in another of your workspaces and Parker opens it in the right one, switching for you.
When you edit a draft an automation wrote, Parker notices what you changed and applies it next time. Each lesson is scored by what actually happened to the draft it produced, so guidance that helps is kept and guidance that does not is dropped. You choose what an automated drafter is allowed to remember.
Parker checks a staged draft against your mailbox, so a run's record reflects what was actually sent rather than what was attempted, wherever your mail provider filed it. Each automation shows what became of the drafts it wrote, and every draft is followed on its own schedule rather than waiting for one daily sweep.
If a question is worth asking every week, turn the thread you asked it in into a scheduled routine without starting again. A card in the conversation opens the routine editor directly.
A preview run on the create page shows exactly what a routine would produce and delivers it to nobody, so you can tune the prompt before anyone receives anything. You can also duplicate an existing routine as a starting point.
Add a teammate to a routine and they receive it with an access level, the same way a shared conversation works, rather than as a bare email address. The per-organisation routine limit rose from three to ten.
A shared conversation shows who is currently viewing it, and sharing is visible from the thread itself rather than hidden in a menu.
Open a conversation someone shared with you and Parker states plainly what it contains and what you can do with it, before you type anything.
A threshold can now be four hours rather than a whole number of days, and a sub-day threshold can be corrected after the fact instead of only being read back to you.
Share read-only so a colleague can read the thread and the answers in it, or share as a participant so they can carry it on. You can change someone's level afterwards without sharing again, and each person's level is shown beside their name.
Messages are attributed by person, so a shared thread reads as a conversation rather than a transcript with no speakers. Threads shared with you get their own place in the sidebar with unread state.
The activity log became a footnote rather than a wall in front of the reply, and each step now says what it actually did.
Saved automations are re-audited regularly, so a change to what you can see in Salesforce surfaces before the next run trips over it. Parker also watches for automation sprawl and tells you about it. It only ever warns; it never removes anything.
See at a glance which automations are healthy and which need attention. Retries are aware of each automation's cadence, and a failure states its reason rather than just failing.
An automation can draft into a digest, reply to the correct address on a case or a deal, and watch for fields that have gone stale.
Prompt history, a PDF record of what each run produced, and a delivery audit, all on the routine's detail page.
Parker builds small interactive views inside an answer when they beat prose: a sortable table, a filterable breakdown, a side-by-side comparison. They are saved with the conversation, so an answer from last month still works today.
Process and relationship questions get a real flowchart with a proper layout. Because Parker now decides when a picture helps, the manual Visualize toggle is gone.
Define how records should be labelled, run it across your data, and schedule it to keep running. Settings shows every classification, its progress while it runs, and controls to enable, disable, rerun or reschedule it. Accuracy is measured against a human-verified reference set rather than estimated.
Classifications is in limited trial with selected organisations. Ask your account contact if you would like to take part.
Notifications raised by background work appear the moment they happen, each one links straight to what it is about, and the panel paginates.
Write house rules directly, with no source document required, add regulations from the Compliance tab, and choose whether a rule applies to one agent or the whole organisation.
Rules are applied across what Parker sends, including email, reports, calendar invitations and the notes it writes into your CRM. Your organisation decides whether a rule observes and records, or blocks outright, and every decision is stored with the mode that was in force. When something is held back, the person who can release it is told.
Outbound commercial mail carries a disclaimer configured for each agent, and sender identity settings moved to the organisation level so different teams can be configured independently.
The step-by-step wizard is gone. Describe what you want watched in a sentence, and Parker shows it back as an editable card you can adjust field by field. The detail page shows the same steps you built, so what you see is what runs.
Fields an automation cannot read are caught while you are building it, rather than when it next runs.
Parker proposes automations built from what you can actually watch in your own data, tests each one, and only offers the ones that work.
Add a website and Parker keeps it fresh: scheduled refreshes, detection of what actually changed since last time, a page-by-page view of the difference, and a preview before you commit to adding it. Every source shows how fresh it is, and Parker takes that into account when it answers.
A connectors gallery, opening with Google Business and Google Places in beta. Watch your locations, pull live review snapshots, and let a new review trigger an automation. Parker can answer "what happened last week" across your connected sources.
Filters are checked against your organisation's actual data before running, so a value that does not exist or a field that is barely populated raises a warning before you act on the result. Parker also learns which fields it keeps getting wrong in your organisation and stops repeating the mistake.
The profile Parker builds of how your Salesforce is actually used now refreshes on a schedule, and tells you when what it is working from is getting old.
Where you let an automation draft on its own, the reply is grounded in your organisation's own knowledge rather than general assumptions.
Charts appear while the answer is still being written rather than at the end. Click a bar or a point and Parker offers a follow-up question about exactly that segment. Scatter and shared-axis combination charts were added, crowded categories fold into a sensible top group, and the palettes are validated for colour vision deficiency in both light and dark themes.
A chart you saw on screen appears in the exported PDF as a properly typeset exhibit.
Parker reads your actual mail threads, newest first, with attachments selectable by name and their contents available to the conversation. Compose a reply in chat and it goes into the real thread. Drafts can be saved to your mailbox for you to send yourself.
Move or edit any event, add and remove invitees, cancel with a message, respond to invitations, and get time suggestions that work across several attendees. Changes made elsewhere sync back, so the calendar Parker sees is the one you see. Free slots respect the working hours on your connected mailbox.
Describe what you want watched in plain language and Parker turns it into something that runs on a schedule. Three shapes cover most work: reacting to each new record or thread as it arrives, ranking a batch into a digest that asks for your consent row by row, and watching for a threshold to be crossed, then resetting once it clears.
Automations can watch Salesforce, draft outreach, create tasks and write back through a staged view that shows exactly what would change before it changes. Where your organisation allows it, an automation acts on its own inside a budget you set, with a full audit of everything it did.
Upload an .xlsx workbook as a data source and query it in conversation. Spreadsheet attachments keep their structure when Parker reads them, and scanned PDFs are transcribed in the background rather than holding up your question.
A per-organisation style setting enforces length limits, regional spelling and your own conventions, so an exported report reads the way your team writes.
Tables were rebuilt in a quiet ledger style with proper headers, sensible column widths, a sticky header on long tables and a share-as-image action. Text and tables share one measured column width, and there is a full dark theme.
When a question would be better answered on a schedule, Parker proposes turning it into a routine there and then.
Push notification, email or in-app, picked when you set the routine up.
Parker tells you in the conversation the moment it learns something, and says what kind of thing it learned. A memory can be pinned, scoped or deleted outright, and an Agent notes surface shows the whole store. An organisation can turn learning off entirely.
Navigation regrouped into Account, Workspace and Data. Agent setup became a single page in briefing order, with advanced options folded away. Every connection appears as a row in one data sources catalog, and organisation knowledge and memories moved under Data.
Paste rough notes about how your agent should behave and Parker turns them into clean, structured instructions.
An acknowledgment appears immediately, in parallel with the real work, so a complex question no longer opens with silence. A live ribbon narrates what Parker is doing while it does it, and its full reasoning is visible rather than clipped to a single line.
Drag a file anywhere on the page or paste an image straight into the composer. Your draft survives navigating away, pressing Enter mid-answer queues your next message instead of losing it, and any turn can be regenerated or edited and resent.
Upload a spreadsheet and ask questions of it in conversation. An explorer shows column statistics and sample rows so you know what you are working with, and answers cite the source the same way Salesforce answers do.
Record updates and bulk creates require an explicit confirmation, offered as Confirm, Reject and Ignore beside the proposed change rather than in a modal you have to dismiss. Custom fields can be created from the conversation.
Parker profiles how your Salesforce is genuinely used and builds an Org Brief from it, so answers reflect your fields, your naming and your process rather than a generic schema.
Individual claims carry a mark showing how well grounded they are, a review mode shows the whole answer's standing at a glance, and any claim can be challenged and re-run against the source data.
An inline card shows progress as a report is composed, and can be stopped part way if it is not going where you wanted.
Set a workspace cover and logo in Settings and they appear on reports, executive summaries and exported conversation records. A Reports hub replaced the old export dialog.
Parker proposes a plan for a substantial investigation, you review and approve it, and it runs in the background and delivers the finished report into your conversation when it is done. The same report can be set to recur, with each edition aware of the one before it so trends read as continuity.
Turn a conversation into a research-style PDF, written as a document rather than dumped as a transcript. A separate conversation record captures the exchange verbatim when that is what you need.
Answers carry a line reading "Analyzed N of M records", so a figure drawn from a sample is never mistaken for a complete count. Where coverage is partial, Parker offers to go deeper.
Parker's main model moved to Claude Sonnet 5, with a noticeable improvement in multi-step reasoning over CRM data.
Parker reads long free-text fields across many records and returns ranked themes, each linking back to the cases behind it.
If you belong to more than one organisation you can move between them without signing out, and your digests and notifications cover all of them.
Changed and removed. Views and Campaigns were retired from Parker during August 2026 and are no longer available.