Launched this week
Katalyst is the AI sales agent for teams on Salesforce. Hang up a call and it's already done: notes summarized, records created, fields updated, follow-up drafted, next step set. It runs 24/7, reading every call, email, and calendar, surfacing the right signals, prepping you for each meeting, and prompting you to act: follow up here, this one’s slipping. What's new: AI Resolution on every account, meeting recorder, hygiene scores, deal patterns. Built for enterprise sales teams.
Katalyst
Hi, I'm Div, Founder & CEO of Katalyst.
At Datadog, I watched our best enterprise reps closing multi-million-dollar deals - lose hours every week to work that had nothing to do with selling. Updating Salesforce, prepping for QBRs, rebuilding context that already lived in an email or a call recording. The people best at selling were spending the most time not selling.
I couldn't unsee it.
So we've spent the last year building Katalyst to fix it: working hands-on with sales leaders and reps across teams like Stripe, Atlassian, Justworks, Datadog.
Katalyst is an AI sales agent for teams on Salesforce. It runs 24/7, reads your calls, emails, and calendar, and turns it into action:
- Salesforce records created and fields updated on its own
- The right signals surfaced (across 10k, hiring, contact tracking, partnerships)
- Account plans created, contacts researched
- A brief ready before & after every meeting
- The follow-up drafted and the next step set
✨ We've built Katalyst to work with your existing Salesforce set-up!
What’s new:
- AI Resolution on every account: like claude cowork, take deep actions on every deal with the Katalyst Agent
- Pipeline with Kanban view, group by and custom views
- Katalyst meeting recorder - joins meetings across Meet, Zoom, Teams, Webex
- Slack bot - push updates, signals, actions to take
- Actions overview - Review and take pending actions, plus bulk actions
And it's already working. Mike, Sales Director at Aniai, has his reps taking 20% more meetings a day, walking in fully prepped, and moving deals to contract 50% faster - all while Salesforce stays clean on its own.
@mike_conway - "If you're not using Katalyst, you are missing out on opportunities your competitors will find."
We just opened Katalyst to the public. If your team runs on Salesforce, you can sign up or book a demo. Would love your honest feedback. 👇
P.S - A huge shoutout to the tens of beta customers without whose love and feedback we wouldn't be here. Also to the elite Katalyst engineering and product team!
Build Check
@mike_conway @divyansh_lohia you're awesome guys! Congrats on this impressive launch
@mike_conway @divyansh_lohia The Salesforce integration is slick. How did you decide on which pipeline actions to automate first? Did early users ask for specific workflows or did you ship what felt logical?
Katalyst
@mike_conway @clquek Appreciate that, the Salesforce layer took the most work by far. Bit of both, honestly. We came in with a clear guess about the actions reps repeat and resent most, (stage, next steps, close dates, logging call outcomes), then early users shuffled the order on us fast. What we assumed mattered most wasn't what they reached for first
Build Check
I was genuinely excited to hunt this one.
What caught me is the narrative, and I haven't seen it anywhere else in the AI sales space. Katalyst isn't trying to replace your systems. It builds on top of Salesforce. Salesforce becomes the data lake, and reps keep working exactly how they already do. No workflow change. No new tool to learn.
It's also proactive in a way most sales AI isn't. This isn't ChatGPT or Claude where you're constantly feeding it context. And the agentic onboarding is impressive. Super slick, simple and easy.
Everything lives inside Katalyst, and it keeps learning from every interaction the rep actually has. The agent runs continuously in the background and takes action on its own. It doesn't wait to be prompted. It just does the thing.
The big win is killing the admin work reps are buried in. But it doesn't stop there. It catches signals on your largest accounts, builds account plans, drafts pre- and post-meeting briefs, even records the meetings. And it isn't just for reps. Managers get a full pipeline and hygiene dashboard across the whole org: how every deal is tracking, how every rep is performing, how clean the pipeline is. A real 360 view, so nobody's scrambling for deal status.
In a crowded space, this is an impressive build. Mostly I just want this community to try it, because it solves a real pain. Take the admin off reps' plates so they can get back to the human part, talking to customers and closing deals.
Katalyst
@german_merlo1 You nailed the core idea, Salesforce stays the source of truth and reps just keep selling. Katalyst quietly does the heavy lifting in the background so nothing falls through.
Scarlett.
@german_merlo1 Great hunt, German!
@divyansh_lohia How does Katalyst decide when to automatically update Salesforce versus when to ask the rep for approval first?
Katalyst
@german_merlo1 @byalexai Early on it asks and shows its reasoning, the rep approves each write. As it proves accurate on a field, you can let that one auto-write while keeping high-stakes fields on approval. Earned field by field.
The write-back is where messy Salesforce instances actually bite. Reading a 5-year-old config is tractable, but the writes hit validation rules and required-field gates that reject silently, and a background agent that thinks it logged the call when Salesforce bounced the update is worse than no agent, the rep trusts a record that isn't there. How does Katalyst handle a rejected write? Does the rep get told 'this didn't save because of rule X', or does the agent try to repair the payload and resubmit on its own?
Katalyst
@dipankar_sarkar Great question, and it's the exact failure we were paranoid about. A write isn't "done" until Salesforce confirms it, if a validation rule or required field rejects it, the rep sees "didn't save because of rule X," never a false green check. And we surface the blocker rather than silently retrying with a guessed payload, because that's how you get bad data that passes validation. Human stays on the calls that need judgment
Katalyst
@dipankar_sarkar Small addition to Div's answer. A suggestion is only marked done after Salesforce confirms the write. If Salesforce bounces it, nothing gets marked accepted and nothing changes in Katalyst, the suggestion just stays in the rep's queue with the Salesforce error. And since reps can edit a suggested value before accepting, they can fix what the rule complained about and accept again. No repair-and-resubmit happens without them.
Super cool gonna try this !! Does it actually write back to Salesforce after calls or do you still have to touch anything manually? What about if it's a call with a prospect who isn't on Salesforce yet? Also wondering if the whole team sees the same view or if it's per rep. Can we share context among reps or is this more individual?
Katalyst
@ishaan_arora1 It does write back to Salesforce after calls! Any essential information from calls get updated on salesforce automatically. For a prospect who is not on Salesforce, we still track their meetings and emails. Any communication from those are on the platform. For views, it actually depends. You can choose to view your own opportunities, or the entire teams opportunities as well!
Katalyst
@ishaan_arora1 To add on the team question, the pipeline data is shared so everyone works off one source of truth, but each rep's agent also carries its own context from their calls and emails. So you get shared visibility across the team without everyone's view turning into noise.
Reading calls, emails, and calendar to keep opportunities current is the right wedge — that hygiene work is exactly what eats selling time. My operator question is about the write path's identity: does Katalyst write to Salesforce as each rep with their own permissions and sharing rules, or through one service account that can quietly sidestep them? And on a multi-threaded deal where two reps' calls touch the same opportunity with conflicting next-steps, how does it decide whose input wins?
Katalyst
@hazy0 Nothing gets clobbered silently. Both reps' inputs land as attributed activities; the canonical Next Step reflects the most recent call, but the conflicting one stays visible and both reps get flagged, so a human resolves it instead of the system guessing
@divyansh_lohia This is the part I was hoping to hear â attributed activities plus a human resolving the flagged conflict beats the system guessing. The half still open for me is the write-path identity: does Katalyst push those Next Step and activity writes into Salesforce as each rep (respecting their permissions and sharing rules), or through one integration user? In a customized org with validation rules and field-level security, writing as a service account is what tends to quietly bypass the guardrails admins lean on.
I've set up and managed a few Salesforce instances over the years, and they're rarely clean 😅. Between custom objects, validation rules, custom fields, page layouts, and all the org-specific logic, introducing a new tool can be a little nerve-racking.
How does onboarding usually work with Katalyst in a more customized Salesforce environment? And for the reps, what's the learning curve like? In my experience, even great tools struggle if AEs have to change how they work too much.
One other thing I'm curious about: can teams add their own AI-enriched fields or custom columns (like account insights, deal risk, buying signals, etc.), or if everything fixed and connected only with salesforce instances?
Katalyst
@kshirja_diwan Great questions. Onboarding is agentic, not a config project, so Katalyst reads your real org and writes inside your existing permissions and validation rules rather than bulldozing your setup. The rep learning curve is close to zero, since it runs in the background off calls and emails and keeps Salesforce current for them, so AEs never change how they sell. And the enriched signals like deal risk, buying signals, and hygiene write straight back into Salesforce as real fields you can use
question nobody's asked yet - what's the correction path when the AI writes something wrong to Salesforce? if it misreads a call and updates a field incorrectly or logs the wrong next step, does the rep even notice before it compounds, since the whole pitch is that they stop double checking. an audit trail is different from a rep actually catching an error in real time
Katalyst
@galdayan two honest parts to the answer. First, the checkpoint sits before the write, not after: early on a rep isn't double-checking a finished record, they're approving the update with its reasoning shown, so a misread gets caught at the moment of decision, not discovered later. Trust is earned per-field, so the fields a rep lets it write unattended are the ones it's already proven on, and the higher-stakes ones (closed-won, forecast category) can stay approval-only indefinitely. Second, the honest part: no system that ever writes autonomously can promise zero missed errors, which is exactly why we scope autonomy narrowly rather than flipping everything to auto and hoping.
The goal isn't a rep who blindly trusts, it's a rep who only stops checking the things that have earned it.
Katalyst
@galdayan on the "does the rep even notice" part specifically: nothing the agent writes is silent. Every autonomous write pings the rep the moment it happens, in Slack and in the app, saying exactly which field changed and to what, with a link to the deal. So it's not an audit trail you dig through later, it's a nudge in real time. And every change keeps the old value next to the new one, so if something does look off, reverting is one edit, not detective work.
@kritikajalan29 that actually closes the gap for me - the real-time nudge plus keeping old/new value side by side is the part that matters, since it turns "did the AI mess up" from a question you have to go looking for into one it surfaces itself. good answer