How to build an AI-native sales org, by Anthropic

What happened at Anthropic, and why it forced an AI-native sales org

Most advice on how to build an AI-native sales org is written by people who have never had to. This one is not. Eleanor Dorfman, Head of Industries at Anthropic, gave a talk at SaaStr AI Annual 2026 describing exactly what her team did when Claude demand went vertical over a single winter break, and what broke on the way.

The short version: they had done extensive annual planning, and the plan was obsolete by the second week of January. The Opus release that landed just before the break changed the shape of the market, and the team came back to demand they had not hired for and had not built process for.

Hiring was not the answer, and this is the part worth sitting with.

Even if we were prepared to hire and onboard and 3x or 4x or 5x the sales team, you simply cannot absorb that number of bodies at the pace that we needed.Eleanor Dorfman, Head of Industries, Anthropic

Onboarding capacity, not recruiting capacity, is the real ceiling. So they rebuilt the system instead. What follows is the four investments they made, plus a layer the talk did not cover: where this playbook gets harder if you are not Anthropic, and which piece to copy first if you have one quarter and no AI team.

The four constraints that forced the rebuild

Dorfman framed the whole rebuild around four things she could not move.

1. Demand she could not staff against. Recruiting standards were not going to be lowered to hit a number.
2. An existing stack she had spent three years building around. Ripping it out was not on the table.
3. Supporting functions that had to scale in lockstep. Sales does not close deals alone. Legal, deal desk, RevOps, billing, compliance, and support all had to meet the same moment.
4. AE capacity that had to grow without burning the team out.

Every design decision that followed is downstream of those four. If your constraints are different, copy the reasoning, not the architecture.

Investment 1: Kill the PLG versus SLG divide

The orthodoxy Dorfman threw out first was the one she had personally defended for years: product-led growth and sales-led growth are separate motions run by separate teams, and enterprise plans get gated by a human.

Anthropic launched an enterprise self-service MVP in January 2026 and pushed it to production in February. Leads are enriched and qualified by Clay and Claude, then routed into one of two funnels. Self-serve buyers get guided by Intercom's Fin, which Anthropic partnered with to repurpose from a support product into a sales product. Sales-qualified leads go to the BDR team and get AE-routed.

The number that matters: 54% of Anthropic's new enterprise logos in 2026 came through the self-serve funnel. These are real ACV deals with terms of service, invoicing, provisioning, and onboarding enrollment, all completed without an AE.

That tracks with where buyers already are. Gartner found in March 2026 that 67% of B2B buyers prefer a rep-free buying experience, based on a survey of 646 buyers, with 45% using AI during a recent purchase.

The reframe is the useful part. Dorfman does not treat self-serve as a downgrade or a consolation prize for small logos. She treats it as friction removal: the right buyer, on the right plan, at the right time. If a buyer wants to swipe a card for an enterprise plan at 11pm, making them book a discovery call is not qualification. It is tax.

Investment 2: Claude as connective tissue, not the seventh tool

This is the load-bearing idea of the whole talk, and the one most likely to get missed.

Anthropic's lead-to-close architecture runs on six tools: Clay for enrichment, LeanData for routing, Salesforce for opportunity management, Gong for call coaching, Ironclad for contracts and redlines, and Slack for deal coordination. Snowflake, BigQuery, Jira, Greenhouse, and G Suite sit around the edges.

It is not a new stack. It is being intentional about your existing stack.Eleanor Dorfman, Head of Industries, Anthropic

Claude is not tool number seven. Claude is what makes the six talk to each other. Context flows from Salesforce to Slack to the AE's head and back into Claude, and Claude fills the gaps between systems that nobody ever built an integration for.

What that looks like on a Tuesday

• Lead lands. Claude and Clay do account research, prioritization, and record updates before an AE touches it, pulling historical context from Slack, Google Docs, and previous Gong calls.
• Every customer-facing rep starts the day in Claude. The morning brief runs across Gmail, Gong, Slack, Google Docs, Calendar, Salesforce, Intercom, and Greenhouse. Dorfman has hers delivered to Slack at 7:00am ET. It surfaces what is on the calendar, which emails are unanswered, which Slack threads need action, which central initiatives are outstanding, and which three things to do first.
• Proposals get drafted by Claude. Instead of an AE opening nine tabs of deal desk guidance and scrubbing Gong transcripts for negotiation history, Claude drafts the proposal against policy and precedent, checks it against customer expectations, then uploads it into Ironclad and kicks off the process.
• Coaching is dynamic, not quarterly. Claude surfaces six coaching moments a week for AEs and frontline managers, generated from what the business needs this week rather than from a static methodology deck. In a market where competitive positioning shifts weekly, a battle card refreshed each quarter is a liability.

This is the deeper answer to the productivity problem the industry keeps restating. Salesforce's State of Sales research found reps spend just 28% of their week actually selling, with the rest going to deal management and data entry, across 7,775 sales professionals in 38 countries. Adding another point tool has never fixed that. Removing the connective work between tools might.

Investment 3: Make Slack the front door

This is the section nobody else writes about, and it is the one I would steal first.

Before the rebuild, getting a quote approved or a security questionnaire cleared meant a sea of DMs and institutional knowledge. AEs on the East Coast and in Europe were staying up late chasing approvals. Some reps just walked past the RevOps desk. That is not a process. That is proximity as a routing algorithm.

Anthropic inverted it: reps stopped going to the systems, and the systems came to the reps. Slack became the single front door for deal desk, legal, RevOps, and support.

Slack in, ticket out. Claude triages and either resolves the request against precedent and policy or escalates it into Jira with the customer contacts and the full history attached, assigns it to a human, and tells the AE it is queued so they can set expectations with the customer.

Sales leaders are rapidly becoming systems thinkers over deal strategists.Eleanor Dorfman, Head of Industries, Anthropic

Getting deal desk tickets submitted properly is not glamorous work. It was also, in her telling, a massive accelerant. The same discipline applies at closed won, which she describes as a checklist with real consequences. Terms of service in place, provisioning path live, onboarding enrollment triggered, invoice issued, org mapping correct. Anthropic runs one skill per closed-won step, for both AE-led and self-serve deals, because the fastest way to ruin a hard-won customer experience is a billing lockout in week two.

Investment 4: Encode your best reps as skills

The fourth investment is the most transferable. Anthropic documented what its top reps actually do and shipped it as a sales plugin: a bundle of connectors and skills that every rep gets on day one, summoned with a slash command.

The reason is capacity math. Anthropic can no longer model productive ramp the old way, where a rep gets their first account on day seven and their first deal in week six. Reps go through boot camp, get a territory, and get the plugin. This is sales enablement inverted: instead of a curriculum reps consume, it is a set of behaviors the system performs with them.

Two details in there matter more than the list.

First, the human stays in the loop by design. The follow-up skill drafts into the AE's inbox and reminds them the next morning if they never clicked send. It does not send on their behalf. Call prep produces the briefing, and as Dorfman put it, the rep still has to read it.

Second, the skills know the brand. Anthropic is protective of its brand and did not want AEs shipping AI slop to customers. Encoding brand rules into the skill is what makes letting any rep generate custom collateral a safe policy rather than a reputational risk.

What the talk left out, and what I would watch for

Three honest gaps, because a playbook you copy without its failure modes is just a highlight reel.

• Forecasting got harder before it got better. Dorfman was candid that every forecast meeting still opens with at least ten minutes debating how to forecast, and that forecast accuracy is not quite there yet. Claude runs the reconciliation across Salesforce, Snowflake, and consumption data, managers inspect and submit, and the call itself is for discussion rather than status reporting. Automating forecast assembly does not automate forecast judgment.
• Some leads fall between the two funnels. She said it plainly: you might hear some people who feel that they ended up in neither funnel. Any binary router will produce that. Build the escape hatch on day one, not after the first angry email.
• This is easier when your product is the AI. Anthropic's reps use Claude all day and can talk about it credibly on calls because they live in it. If you are running this playbook on top of a tool your team is skeptical of, adoption is your first problem and architecture is your second. Solve them in that order.

One more thing worth naming. Gartner also found that 69% of B2B buyers turn to sales reps to validate AI-generated insights. Buyers want less rep-mediated process and more rep judgment. That is precisely the trade an AI-native sales org is designed to make: automate the assembly, protect the judgment.

How to start building an AI-native sales org this quarter

Dorfman closed with four moves. Here they are, with what I would actually do first.

1. Turn on the AI already embedded in your stack, intentionally. Not we enabled the AI feature in Salesforce. Ask how it connects to the rest of the customer journey, or you will have six disconnected copilots.
2. Thread AI through the sales cycle you already run. It is an accelerant, not a replacement motion. Map the customer experience and the AE experience stage by stage, then find the handoffs where context dies.
3. Make Slack or Teams the front door for one supporting function. Start with deal desk or legal. This is the highest-leverage, lowest-risk change on the list, and it wins you political capital with the teams you will need later.
4. Document what your best reps do and ship it as a skill. Start with call prep. It is the easiest to specify, the easiest to evaluate, and the fastest to show value.

The rule underneath all four: do a thing manually once, then train the system to do it the next time. That is what turns a pile of tools into an AI-native sales org.

If you only do one thing this quarter, do number three. Everyone reaches for the rep-facing copilot because it demos well. The constraint is almost never the rep. It is the eleven-day quote approval sitting behind them.