What Vercel's AI-run GTM tells you about what to staff vs. automate

What Vercel's AI-run GTM tells you about what to staff vs. automate

Tom Occhino, CPO at Vercel, just went on record: 96% of Vercel's marketing runs on AI agents, 93% of support is handled by agents, and their entire SDR team has been reabsorbed into the org. Read the full breakdown in How Vercel Runs on AI Agents: 96% of Marketing, 93% of Support, and an SDR Team Reabsorbed. What jumped out: this isn't a startup hacking together automations. Occhino built React and React Native at Facebook. When someone with 12 years of foundational infrastructure work says the SDR function is gone, that's not a hot take. It's an architectural decision.

The signal for post-PMF B2B SaaS founders isn't that you should fire your SDR team tomorrow. The signal is that the function itself is being redefined. Outbound was always about volume, consistency, and personalization at scale. Agents do all three better than humans for a certain category of motion. What Vercel's data tells you is which motions those are, and by implication, where human GTM judgment still compounds.

The SDR role didn't die. The repetitive version of it did.

Occhino didn't say sales is automated. He said the SDR team was reabsorbed. Those people are still there. They're just not sending 200 templated sequences a week anymore. The work they now do is judgment-heavy: ICP refinement, narrative sharpening, relationship escalation. The research-and-spray motion is gone. If your outbound is mostly high-volume cadences to semi-qualified lists, an agent can run that motion at 1/10th the cost and 3x the consistency. The question is whether that motion was generating pipeline or just activity.

96% of marketing is a channel signal, not a headcount signal.

The number that matters isn't the percentage. It's the implication: content generation, distribution, and performance analysis can run continuously without human bottlenecks. That means the founders who win aren't the ones who hire the best content team. They're the ones who build the tightest GTM strategy loop: identify the signal, generate the asset, score the result, iterate. Agents compress that loop from weeks to days. If your current marketing motion takes 3 weeks to go from idea to published to analyzed, you have 15 days of unnecessary drag that compounds every cycle.

Support at 93% agent-handled redefines where human CS creates value.

At 93%, the agents are handling everything that has a deterministic answer. That's most tickets. The 7% that goes to humans is exception handling: escalations, nuanced objections, situations where being wrong costs a customer. For B2B SaaS founders, this means your support headcount should be sized for the exception rate, not the ticket volume. If you're staffing based on total tickets, you're overstaffed for agents and understaffed for judgment. The same logic applies to GTM intelligence. GTMVP runs the monitoring layer on agents for exactly this reason: the repetitive scanning compounds when it runs continuously, and the human only steps in on the calls that move pipeline.

The growth motions agents can't replace are the ones worth staffing.

Agents are excellent at tasks with clear success metrics and repeatable structure: outbound sequencing, content generation, ticket resolution, keyword targeting. They're poor at tasks that require reading a room: a enterprise discovery call, a positioning pivot when the ICP turns out to be wrong, a relationship that converts because of trust built over 6 months. The GTM strategy work that actually moves pipeline at Series A is in that second bucket. That's where founders should spend human hours, including their own.

What this means for paid media specifically.

Vercel's numbers don't directly address paid channels, but the pattern extends. Campaign setup, audience iteration, and performance reporting are all agent-appropriate tasks today. The part that isn't: deciding which signal to optimize toward. If your Smart Bidding is trained on form fills and your actual revenue comes from enterprise deals that start with a demo request, no agent fixes that misalignment. That's a judgment call about what conversion event actually predicts closed revenue. Get that wrong and the agent optimizes confidently in the wrong direction. I've run $50M+ in ad spend and seen this exact failure pattern in accounts spending $300K a month.

The deeper point: automation amplifies whatever signal you feed it. A clean signal compounds. A bad one compounds faster. Vercel didn't get to 96% by handing agents a vague mandate. They got there by being ruthless about what each motion was actually supposed to produce. Most founders skip that step. They automate the activity before they've defined the outcome, then wonder why the dashboard looks busy and the pipeline looks flat. The order matters: define the outcome, instrument the signal, then let the agent run. Do it in the wrong order and you've just bought a faster way to be wrong.

What to do this week

• Audit every human GTM task in your org and categorize it: does it have a clear success metric and repeatable structure (agent-appropriate), or does it require judgment and relationship context (human-appropriate)?
• Map your outbound motion and identify what percentage of sequences are templated cadences vs. genuinely personalized outreach based on account-specific signals. That percentage is your automation ceiling.
• Score your support tickets for the last 30 days. Separate them into deterministic answers vs. judgment calls. The deterministic ratio tells you your agent-ready support volume.
• Identify the one paid channel where your attribution is cleanest (meaning you know what conversion event predicts closed revenue). That's where agent-assisted optimization can run without poisoning signal quality.
• Check whether your GTM motion has a feedback loop shorter than two weeks. If not, you're iterating too slowly for an agent-assisted model to compound. GTMVP's eight agents run continuous competitor mapping, channel scoring, and trend surfacing precisely because weekly cadence beats monthly in fast-moving B2B categories.

GTMVP exists to run the intelligence layer that feeds those judgments. The eight agents handle competitor monitoring, positioning analysis, angle generation, channel scoring, and trend detection continuously, so the human judgment work starts from current data instead of last quarter's assumptions. If you want to see what that looks like applied to your category, start with the GTMVP audit or pull a sample report to see the output format before committing.