• By Dan Mintz. 3x founder, Wharton MBA, MIT MS in Machine Learning.

• Leading enterprise AI transformation expert with experience in many projects, operating in the intersection of business strategy and AI technologies to drive impactful results.

• The blueprint: business strategy should drive the AI transformation.

Introduction

At SaaStr AI Annual in May 2026, Eleanor Dorfman, who leads Anthropic's commercial and industries sales team, walked through how her team rebuilt its sales org in the first weeks of the year (watch the full talk). The trigger was simple. After a new Claude release, the team came back from the 2025 winter break to enterprise demand that had gone vertical. They had not hired for it, and their processes were not built for it.

Most coverage of the talk focuses on the headline numbers. The more useful lesson for enterprise leaders is the operating logic underneath them. Anthropic did not buy an "AI-native sales platform." It did not rip out Salesforce. It redesigned the workflow, kept its systems of record, and placed an AI orchestration layer between them.

That is the same pattern that separates enterprise AI transformations that produce business impact from those that stall in pilots. This post breaks down what Anthropic did, why it worked, and which parts transfer to a Fortune 1000 company with a messier stack, heavier governance and a slower-moving workforce.

What problem forced Anthropic to rebuild its sales org?

Demand grew faster than any hiring plan could absorb. Dorfman's point was that even a team prepared to triple or quintuple headcount cannot onboard that many people fast enough to protect the customer experience. Anthropic also refused to lower its recruiting bar to close the gap.

The team defined four constraints it could not move:

• Demand it could not staff.

• Claude was already embedded inside the tools it had bought.

• Sales depends on surrounding functions: legal, deal desk, RevOps, billing, compliance, support and customer success.

• AE capacity had to grow without burning out the existing team.

This framing matters more than any single tool decision. Anthropic did not start with "where can we use AI?" It started with a binding business constraint, a capacity gap between demand and headcount, and worked backward from it. That is the order enterprise transformations should follow. Business strategy defines the problem. AI is one of the levers for solving it.

Why did Anthropic keep its existing sales stack instead of replacing it?

Because the stack was the asset. Anthropic had spent three years configuring its tools around its own needs, and they worked. The lineup is familiar to any B2B operator: LeanData for routing, Clay for enrichment, Salesforce as the system of record, Gong for call intelligence, Ironclad for contracts, Jira for tickets, Intercom's Fin for support, Snowflake and BigQuery for data, and Slack and Google Workspace for communication. SaaStr's breakdown of the stack called the lineup the most "boring" part of the presentation (SaaStr on Anthropic's GTM stack).

The decision was to double down on that investment and thread Claude through it. In Dorfman's framing, Claude is not a seventh tool bolted onto six others. It is the connective tissue that lets the existing tools share context and hand work to one another. Technically, much of this runs through connectors built on the Model Context Protocol, the open standard Anthropic created and later donated to the Linux Foundation's Agentic AI Foundation (Anthropic announcement).

For enterprise leaders, this is the most transferable idea in the talk:

• Systems of record stay authoritative. Salesforce remains the source of truth. An AI-generated summary or an agent's memory never becomes the record.

• AI shows up where people already work. Reps did not learn a new application. The work came to them in Slack, email and Salesforce.

• Reuse beats rebuild. McKinsey's Rewired makes the same case: scaling depends on reusing data, integrations and components rather than rebuilding them for every use case.

I applied the same principle in a GTM redesign I mapped for Consensus, a B2B SaaS presales company. The architecture deliberately rejected a custom interface. AI output landed on the existing CRM deal record and in Slack, because adoption tends to die the moment employees are asked to open yet another tool.

Can enterprise buyers really purchase without talking to a salesperson?

Yes, when the funnel is designed for it. Since Anthropic launched enterprise self-service in early 2026, 54% of its new enterprise logos have come through the self-serve funnel (SaaStr's write-up of the session).

Getting there required breaking an orthodoxy that shaped fifteen years of enterprise software: product-led growth and sales-led growth are separate motions run by separate teams, and enterprise plans must be gated by a human. Dorfman had operated by that rule herself for most of her career. Anthropic dropped it. An MVP launched in January and moved to production in February.

The mechanics:

• Every lead is enriched, evaluated and qualified by Clay and Claude.

• Leads then route to one of two funnels. Self-serve buyers are guided by Intercom's Fin, retooled from a support product into a sales tool, and receive a real enterprise plan: real ACV, terms of service, invoicing, provisioning and onboarding.

• Leads routed to sales go to BDRs, then to AEs. The gate shown on the slides was explicit: sellers work only the deals where a seller changes the outcome.

This is workflow redesign, not automation. Anthropic did not make AEs faster at deals that never needed an AE. It removed the seller from those deals and concentrated human effort where it changes the result. SaaStr later made the same point: self-serve is not an AI account executive, it works by removing the need for a closer, and Anthropic still listed 72 open sales roles in late May 2026 (SaaStr on AI agents in sales).

McKinsey's research points the same way. In its State of AI survey, the small group of AI high performers, about 6% of respondents, are nearly three times as likely as others to have fundamentally redesigned individual workflows, and that redesign is one of the strongest contributors to business impact of all the factors tested (McKinsey, The State of AI).

What does an AE's day look like in an AI-native sales org?

It starts in Claude, not in the CRM. Every customer-facing rep begins the day with a Morning Brief built on connectors across Gmail, Gong, Slack, Google Docs, Calendar, Salesforce, Intercom and Greenhouse. It surfaces the day's meetings, unanswered emails and Slack threads, open internal initiatives and upcoming events, then ranks the actions that matter most. Dorfman receives hers in Slack at 7:00 a.m. Eastern. The presentation slides put the impact at 10-15 hours of prep time returned per AE per week.

The rest of the deal cycle follows the same pattern:

• Account research. When a lead becomes an account in Salesforce, Claude and Clay research and prioritize it and pull historical context from Slack, Google Docs and past Gong calls before the AE is assigned.

• Proposals. Claude drafts the proposal against published deal desk policy, historical precedent, negotiation history from Gong and email, and product knowledge, then uploads it to Ironclad and starts the contract process. The slides show draft time falling from 45 minutes to 4.

• Forecasting. Claude reconciles Salesforce against actual spend, commits and customer communications. AEs submit through Claude, and managers inspect and review. Forecast calls become discussions about where help is needed rather than interrogations. Dorfman was candid that this is still a work in progress.

• Coaching. Gong, Snowflake and Claude condense roughly 40 hours of calls into six coaching moments per week, re-prioritized as business needs shift from month to month.

• Closed Won. One Skill per step handles terms of service, provisioning, onboarding enrollment, invoicing and org mapping, for sales-led and self-serve deals alike.

The common thread: the AE stops being the person who assembles, records and transports information through the pipeline. In the Consensus analysis, AEs spent roughly two hours of an eight-hour day actually selling. Research, CRM administration, scheduling and follow-up consumed most of the rest. That is exactly the capacity Anthropic reclaimed, and it is the core of pipeline economics: every hour moved from administration to customer work changes conversion math downstream.

Why do the functions around sales matter as much as sales itself?

Because deals stall in the handoffs, not in the pitch. Before the redesign, getting a quote or an approval at Anthropic meant a sea of direct messages, relying on institutional knowledge, or physically walking over to the deal desk. Reps in Europe and on the US East Coast were staying up late chasing approvals.

Anthropic made Slack the front door for deal desk, legal, finance and RevOps:

1. A seller posts the request in a dedicated channel.

2. Claude triages it against published policy and precedent, and either resolves it or escalates it.

3. Escalations reach a human as a Jira ticket with customer context from email, Salesforce and Gong, relevant precedent and a pre-drafted response. The AE sees their place in the queue.

The design respects the dual role of these functions. They enable deals, and they protect the company. In a large enterprise, that second role raises the bar considerably. The limits on what an agent may resolve on its own must be enforced by the orchestration layer and tool permissions, not written into a prompt. Human approvals must be bound to the exact change being approved, not to a summary of it. These controls separate a working triage system from approval theater, and they are where most regulated enterprises will spend their real effort.

How do you turn top-performer behavior into the organizational baseline?

Encode it as Skills and ship it to everyone. Anthropic could not afford multi-week, handheld ramps for new hires. New reps go through boot camp, receive a territory, and receive a Sales Plugin: a bundle of connectors and five Skills that capture what the best reps already do.

• Morning Briefing: daily prioritized actions across calendar, email, Slack and Gong.

• Call Prep: a one-page brief generated before any meeting, covering attendees, what they care about, deal history, positioning and recommended discovery questions.

• Customer Follow-Up: action items extracted from calls, emails and CRM notes, with drafted responses a human reviews and sends, protecting a 24-hour follow-up commitment.

• Competitive Intel: deal-specific battle cards and comparison matrices instead of static quarterly ones.

• Create an Asset: custom one-pagers, landing pages, ROI calculators and prototypes, with brand guidelines enforced automatically.

Skills are folders of instructions, scripts and resources that an agent loads when a task calls for them, and the format is now an open standard (Agent Skills specification). Their business significance is far larger than their technical simplicity. They convert tacit tradecraft into a versioned asset that can be distributed, audited and improved.

This addresses one of the most expensive problems in sales. In the Consensus case, roughly the top 14% of AEs generated about 80% of revenue. The largest prize was not making the stars slightly better. It was moving the middle of the distribution closer to the top. Encoding what top performers do, and making it the default for every rep from day one, is how that happens at scale.

What should enterprise leaders copy, and what won't transfer directly?

Copy the logic. Do not assume the conditions. Anthropic's own four recommendations are sound:

• Turn on AI where it is already embedded in your tools, deliberately, and tie it to one coherent customer journey.

• Thread AI through the sales cycle you already run rather than inventing a new one.

• Make Slack or Teams the front door for support functions.

• Document what your top reps do and ship it as Skills.

A Fortune 1000 company should also be honest about the differences:

• Anthropic is the vendor. Its workforce is AI-fluent by default, and model access is a given. Most enterprises face procurement cycles, security reviews and change management that Anthropic did not.

• The stack was clean enough. Three years of deliberate configuration around a single system of record is not the norm. Many enterprises run multiple CRMs, regional ERPs and unresolved customer identities. An AI layer that connects bad data produces confident errors faster.

• The numbers are self-reported. The 54% figure, the hours saved and the proposal times come from Anthropic's own presentation. They are credible directional signals, not audited benchmarks.

• Adoption is not automatic. Rewired describes IBM's AI development rollout, where adoption was initially patchy and people reverted to old methods until a serious enablement program followed. Top performers invest in hands-on upskilling and coaching, not passive training.

The practical sequence for a large enterprise follows directly. Treat sales, or another revenue workflow, as a bounded transformation domain. Map it end to end. Challenge each assumption before automating it, starting with questions like "why must the AE research the account?" and "why must the AE enter the data?" Establish a human baseline, pilot against a control group, and scale only after value, risk and platform readiness are each proven.

What is really at stake

The competitive gap in enterprise AI is not access to models. Every large company can buy the same models and the same connectors. The gap is between organizations that redesign how work flows and those that bolt AI onto the old flow and wait for productivity to appear.

Anthropic's case shows that the redesign does not require a new stack. It requires a clear business constraint to solve, a leader willing to retire long-held orthodoxies, and the rigor to treat the space between systems as the place where value is created or lost. Dorfman described the shift as sales leaders moving from deal strategy toward systems design. That shift is now arriving in every function that runs on handoffs, approvals and scattered context, which in a large enterprise is most of them.

Frequently Asked Questions

1. What is an AI-native sales org?

In Anthropic's model, it is the existing sales stack with an AI layer doing the work between the tools: research, drafting, reconciliation, triage and follow-up. It is not a new platform, and humans stay on the deals where they change the outcome.

2. What is the difference between an MCP connector and a Skill?

A connector gives the AI access to a system such as Salesforce, Gmail or Gong. A Skill is a packaged procedure that tells the AI how to perform a specific workflow, such as call prep or a Closed Won checklist, using those connections.

3. Did self-serve replace Anthropic's sales team?

No. Self-serve absorbed the deals that never needed a seller, which accounted for 54% of new enterprise logos in 2026, while complex deals still route to BDRs and AEs and Anthropic kept hiring for sales.

4. Where should a large enterprise start with this approach?

Making Slack or Teams the single front door for deal desk, legal and RevOps requests, with AI triage before human escalation, is usually the lowest-risk start. It sits on existing tools, removes visible friction, and keeps humans accountable for every non-standard decision.

5. How does an enterprise apply Anthropic's approach to its own go-to-market organization?

Start from the business constraint, not the tool: the capacity gap, cost-to-serve, or conversion problem the GTM strategy has to solve. Then map the revenue workflow end to end, decide where a seller changes the outcome, and thread AI through the systems of record you already own. My work sits at that intersection of business strategy and AI deployment, so each change is tied to the competitive advantage it is meant to create.

6. How should leaders plan when AI capabilities change this fast?

Anchor the plan on business objectives, which change slowly, and treat tools and workflows as the part that iterates. Anthropic re-prioritizes its coaching focus month to month because the business shifts that quickly. A transformation roadmap should work the same way: a stable strategic direction, with short cycles to pilot, measure, and scale individual workflows as capabilities improve.

7. Our company has already rolled out AI tools. Why aren't we seeing results?

Tools do not change results on their own, as IBM's early rollout showed when people reverted to old habits without hands-on support. Results come from redesigning the workflow around the tool, tying it to a business metric leadership is accountable for, and enabling people to work the new way. Connecting those three is the core of a strategy-led AI transformation, and it is where most tool rollouts stall.