Rafi (@halim__rafi) header
Rafi (@halim__rafi) profile picture

CTO @octolane · building Superintelligence

San Francisco, CA
Joined July 2023
Photo by Rafi (@halim__rafi)Photo by Rafi (@halim__rafi)Photo by Rafi (@halim__rafi)Photo by Rafi (@halim__rafi)Photo by Rafi (@halim__rafi)Photo by Rafi (@halim__rafi)Photo by Rafi (@halim__rafi)Photo by Rafi (@halim__rafi)Photo by Rafi (@halim__rafi)Photo by Rafi (@halim__rafi)
The best thing my co-founder has ever written, and he wrote it from a café table while we hit profitability. We bet the entire company on the argument in this essay. Tell him where he's wrong. He loves that.
Every lab is fighting over the same scraped internet. Meanwhile the highest-value training data gets destroyed daily because nobody's positioned to catch it: an expert saying yes, fixing it, or killing it, with the outcome attached. We are. Argue with me.
2
1
10
883
Shared inboxes are where teams work together. Your agents should be able to work there too, with the right permissions. Now they can in @octolane 🚀 Updates here: octolane.com/updates#shared-…
Just shipped: Your agents now have shared mailbox access and role-based permissions 🚢 Here's what that means in @octolane - Self-driving AI CRM for Agents: → Agents can send from shared mailboxes. If a teammate shares their inbox with you, your agent can work it too: follow-ups, replies, outreach, from the right address. → Agents work within your role. Owner, Admin, or Member, your agent gets exactly your permissions. Nothing more. → You control who can send from each mailbox: only you, all members, or the teammates you pick. Your agents follow the same rules. → Shared mailboxes sit right beside your own. Set one as your default and you and your agent send from it automatically. Your team shares inboxes. Now your agents can too, safely.
1
1
5
221
I learned this at @ycombinator W24 batch, here PG says it beautifully in just 3 sentences. School trains us to study the grader. A startup is the first game where the grader never shows up. A user opens your product or they just don't. Consider Stripe. In the early days, when someone agreed to try it, instead of sending just a signup link the Collisons said "give me your laptop" and set it up right there. @paulg later gave it a name but what it really was 2 founders who refused to be anywhere except next to the person using their product. You see the same pattern in @resend and @linear. Founders who were their own most demanding users first, and who kept asking questions like if I were in their shoes, would I actually use this? Would I pay for it? Investors can be persuaded. Users can't. They just leave. It doesn't change as we build for AI agents, either. Every agent works for someone. At the end of the chain there's always a human being with a problem. Which is why this work asks for selflessness. Capital is a tool and investors are a lagging indicator. You need to care about the people you're building for, pour everything into that and the rest follows
The most consistent mistake I see young would-be founders make is to care too much about investors' opinions and not enough about users' opinions. They've been trained their wholes lives to seek out and please the gatekeepers. But this is where gatekeepers stop mattering.
9
1
12
990
A founder who's already won once, choosing Day 1 again, is a dangerous thing for every competitor. Good to have you back in the arena @bhalligan. Let the fight begin ⚔️
I'm back as CEO of @HubSpot today. @dharmesh and I are completely re-founding the company. Its day 1. We're now the Customer Harness Company. In the next few days, I'll post about why I'm coming back and what my plans are. LFG!
1
1
17
1,070
We asked our agent how many API endpoints @octolane has. It found two docs pages, counted the lists, and answered [N]. Both links were real. The real number is [M]. Neither page had the full inventory, and nothing in them showed what was missing. Search found relevant pages. It couldn't tell us whether they were enough. Thanks to @mintlify, getting our docs into the agent's knowledge base was the easy part. The failure wasn't access. It was knowing when the pages weren't enough. So we changed the harness, not that one question. Counting questions now start from a page index, and the agent is told to say when the evidence isn't enough. Not solved: in a later test the count was right, but another run's breakdown didn't add up. Citations tell you the agent found something. Only reading the source tells you it found enough. halimrafi.com/writing/agent-…
1
114
Octolane is the best sales tool ever (not only a CRM)!
We trained a custom model to kill the only moat CRMs have, switching cost. @RespanAI asked @octolane to migrate them off Attio to Octolane. Octolane migrated years of their data in under 4 minutes. That’s Exodus 1, our in-house migration model. It reads your old CRM’s schema, maps every custom field and object, carries over full conversation history, and enriches every account and contact in flight. Attio customers now migrate in under 5 minutes. HubSpot customers in under 20. And we’re already training Exodus 2 to migrate Salesforce customers in under 30 minutes.
1
1
12
453
Why this makes us unstoppable Lock-in was the incumbents’ best feature. Every company that hates its CRM stays because leaving is terrifying. That fear is the whole moat. Exodus 1 turns their moat into our acquisition channel. Every messy, customized, 5-year-old CRM instance is a customer that couldn’t leave until now. And the incumbents can’t answer. No CRM will build a better exit door out of its own product. But why does CRM migration need a model? Because no 2 CRMs look alike on the inside. Every company bends its CRM until it’s unrecognizable. A field called “Status” means lead status in one workspace and churn risk in another. “ARR_final_v2” sits next to “ARR (do not use).” One pipeline has 4 stages, another has 19, and three of those are named “Closed.” Someone built a custom object called “Partners” in 2022 that half the deals still point to. The structures don’t line up either. Attio thinks in objects, lists and attributes. HubSpot thinks in properties, pipelines and labeled associations. Salesforce splits people into Leads and Contacts, hangs deals off Accounts, ties people to deals through contact roles, and attaches activity to records that can be several different types at once. Moving between them is translating one company’s mental model of its business into another system without losing meaning. The hardest part is the history. A CSV export flattens everything. Contacts lose their companies. Deals lose their emails. Notes lose their authors. A 3-year relationship with your biggest customer turns into a spreadsheet row with no context. Exodus 1 rebuilds the graph: every email, note, meeting and call reattached to the right person, the right company and the right deal, with duplicates merged along the way. Then it enriches every record before it lands, so you arrive with better data than you left with. Every migration teaches Exodus 1 a new edge case. The more we move, the faster and cleaner the next one gets. That’s why it’s Attio today, HubSpot daily by February, and Salesforce next.
We trained a custom model to kill the only moat CRMs have, switching cost. @RespanAI asked @octolane to migrate them off Attio to Octolane. Octolane migrated years of their data in under 4 minutes. That’s Exodus 1, our in-house migration model. It reads your old CRM’s schema, maps every custom field and object, carries over full conversation history, and enriches every account and contact in flight. Attio customers now migrate in under 5 minutes. HubSpot customers in under 20. And we’re already training Exodus 2 to migrate Salesforce customers in under 30 minutes.
5
2
13
1,275
🚀🚀🚀
Migrated 2 customers from HubSpot to @octolane so far in the 60m
49
Migrated 2 customers from HubSpot to @octolane so far in the 60m
At HubSpot Ignition right now. Walked straight into the lion’s den. Meanwhile, at @octolane: → we migrate a company off HubSpot every week → we migrate a company off Attio every single day How? We trained our own model for it. Exodus 1 moves your entire CRM over in under 20 minutes. A year ago, people laughed and said no one would ever switch CRMs. Investors passed and wrote us off as too young. Who’s laughing now? Here’s what I’ve learned about winning a market the giants own: 1. Kill the switching cost. Nobody stays on their CRM because they love it. They stay because leaving hurts. So we made leaving take 20 minutes. 2. Go where your customers already are. Even if that’s your competitor’s conference. I’m all in. If you’re at Ignition and tired of your CRM, come find me. I’ll migrate you before the next session starts.
1
1
1
4
622
We started by giving Octo a name and a way to show who it was listening to. Now you can decide how your notetaker shows up: its name, branding, background, and motion. Your @octolane notetaker should feel like part of your team 💪💪💪
Just shipped: @octolane AI notetaker now joins calls as a member of your team When an AI represents your company, whose company should it look like? Yours. So we shipped the customization → Your logo on screen. No upload, it already knows your workspace. → Your name. It introduces itself as your company’s AI, or as yours. → Your look: light or dark, rays, shapes, motion. And Octolane is catching buying signals live, learning how your deals move, and turning every call into the next action in your pipeline. Superintelligence for revenue starts by showing up as you. Live now for every @octolane customer today
1
1
5
210
When people spend hours in a product, the small details matter a lot more than we think: what needs attention, what’s running, and where to go next. The care in @octolane’s design shows up in those everyday moments. Fantastic work, @vitorviesi 👏❤️
Today our top users spend more time in @octolane than in ChatGPT or Claude. @karrisaarinen and the @linear team raised the bar for what software should feel like. We hold @octolane to the same one. A lot of this is @vitorviesi. Thousands of iterations later, people love Octolane. AI made shipping cheap, but caring is still expensive. ❤️
2
4
282
It’s incredible watching the @octolane team build in real time.
Today I showed my team something I wasn’t sure I should post publicly. We trained our own models to close deals end to end. We built our own eval, Deal Benchmark 1, because no public benchmark measures whether an AI can actually close a deal end to end. Our models match frontier models (GPT & Claude) on it, at a fraction of the cost and latency. 4 models for one job. Antenna 3: catches buying signals in real time Driver 4: predicts the next best action on every deal Inkwell 2: trained on world-class objection handling and negotiation Exodus 1: migrates your entire CRM. ~5 min from Attio, ~20 from HubSpot. Next up: Exodus 2 targeting Salesforce. I’ve used 40+ CRMs (including the new hot ones). Every one is a database with an “Ask AI” button, waiting for a human to do the actual work. 1 rep → an always-on agent on every account. When I finished the presentation, the room was quiet for a second. In 12 months, the best closer on your team won’t be a person. API access to all four opens February 2027. Waitlist is open now. Existing customers get in first.
3
4
254
Today I showed my team something I wasn’t sure I should post publicly. We trained our own models to close deals end to end. We built our own eval, Deal Benchmark 1, because no public benchmark measures whether an AI can actually close a deal end to end. Our models match frontier models (GPT & Claude) on it, at a fraction of the cost and latency. 4 models for one job. Antenna 3: catches buying signals in real time Driver 4: predicts the next best action on every deal Inkwell 2: trained on world-class objection handling and negotiation Exodus 1: migrates your entire CRM. ~5 min from Attio, ~20 from HubSpot. Next up: Exodus 2 targeting Salesforce. I’ve used 40+ CRMs (including the new hot ones). Every one is a database with an “Ask AI” button, waiting for a human to do the actual work. 1 rep → an always-on agent on every account. When I finished the presentation, the room was quiet for a second. In 12 months, the best closer on your team won’t be a person. API access to all four opens February 2027. Waitlist is open now. Existing customers get in first.
4
4
1
15
1,716
Good morning San Francisco! I'm all in. Are you?
11
2
72
5,864
What my inbox looks like these days. Teams switching off Salesforce. Companies that outgrew their CRM. Head-to-head evals against newer AI CRMs, and we’re winning on research accuracy and how carefully our agent handles your data on long-horizon tasks. You can’t buy this. We’ve never spent a single dollar on acquisition or paid marketing. No sales team and no ads. When someone who spent most of their career inside Salesforce writes “I don’t think I’ll be going back,” no ad budget in the world gets you that sentence. Last night I told myself I’d finally get some sleep. I took my laptop to the couch, which is my bed now since the bedroom became our office. “Just to check one more thing.” Shipped a fix. Answered a customer. Shipped another. Jumped into a thread with a team mid-eval. Looked up and it was 3am. I should have been exhausted. I’ve never felt more awake. It never feels like work. It feels like the thing you’d be doing anyway. You don’t burn out doing work that loves you back.
7
2
19
1,359
Replying to @octolane
@octolane is gonna cross $100M ARR in 17 months without a sales team. That’s the point. I hand onboarded 1,000+ companies myself. We have PMF now. Strap in. 🏎️🏎️🏎️
Applied Compute update: - crossed ⁠$100M annualized revenue run rate up from zero 18 months ago - raised $350M at $3.25B led by Elad Gil - 40% of customers use AC for both custom model training + production inference - inference serving 1T tokens a day after launching 12 weeks ago - 2X enterprise customers since last raise - 3X team since January, 85% are researchers and engineers ⁠- Working with customers like Microsoft, Shopify, Harvey, DoorDash, Cognition, Clay, Decagon and others To move even faster we need people who are extremely technical AND extremely practical. If this is you, please reach out.
7
2
1
58
4,221
Switching CRMs as another project for your team is dead. The hard part is moving messy data without losing the relationships and context behind it. That’s what we’re working on at @octolane
We broke HubSpot’s moat. This month on @octolane: - HubSpot teams are switching every week, fully migrated in under 20 minutes - A team moved off Attio in 5 minutes - Another migrated complex, messy Attio data in 8 minutes Zero consultants needed. Octolane's agents did the whole migration. Now we’re training our 4th AI model, Octolane Exodus 1, built to migrate entire companies off legacy CRMs at scale. Who said a team of 5 can’t take on giants?
2
3
7
354
We broke HubSpot’s moat. This month on @octolane: - HubSpot teams are switching every week, fully migrated in under 20 minutes - A team moved off Attio in 5 minutes - Another migrated complex, messy Attio data in 8 minutes Zero consultants needed. Octolane's agents did the whole migration. Now we’re training our 4th AI model, Octolane Exodus 1, built to migrate entire companies off legacy CRMs at scale. Who said a team of 5 can’t take on giants?
7
6
2
72
15,475
I'll send $100 to whoever guesses first what this chart measure, and $200 to whoever figures out what changed in May that bent the curve. btw this is a PostHog chart from @octolane
7
1
5
1,311
just got 6 foot whiteboard for real-time target tracking for the team, plus servers, enough GPUs to take our models to the next level. My bed didn't make the cut, so I'm back on the couch. The whiteboard is blank today. It won't be by Monday. Let’s go @octolane
3
1
8
640
Cleaning up the design for accounts import widget for @octolane 🖌️ w/ @coffeewithone @vitorviesi
8
4
3
27
2,038