Concept

Go-to-Market Engineer

conceptgo-to-marketsalesaiagentic-engineering

Go-to-Market Engineer

A hybrid role that combines sales domain knowledge with engineering skill to build AI agents and automated workflows for go-to-market functions. The GTM engineer encodes the workflow of a top-performing human into an agent, keeps a human in the loop while KPIs are validated, then redeploys the human to higher-value work.

Origins

The role’s conceptual forerunner was Project Rosland at Stripe (2017): an attempt to build a “company universe” database (every row a company, every column a predictive attribute) and populate outbound email templates automatically — a “mad-libs” approach to personalisation at scale. It mostly failed because LLMs didn’t exist and false-positive rates in attribute matching were too high.

The same concept became viable around 2023–24 with the availability of capable LLMs, agentic frameworks, and better data infrastructure. The GTM engineer makes the 2017 vision executable.

The core workflow

  1. Shadow the top performer. Map every tab they open, every data source they consult, every judgment call they make.
  2. Encode the workflow. Build a deterministic + LLM-assisted agent that replicates the process — some steps rule-based, some model-guided.
  3. Human in the loop. A single human reviews every agent output before it is sent or acted on.
  4. Track the baseline KPIs. For an SDR, this is lead-to-opportunity conversion rate and touches-to-convert. For a deal review, it is loss-reason accuracy.
  5. Iterate on rejections and edits. The agent incorporates reviewer feedback as training signal.
  6. Redeploy. When the agent matches human baseline performance, move the freed humans up the value chain.

Worked example — Vercel’s lead agent

Before: 10 SDRs qualifying inbound leads. After: one SDR doing QA on the agent. Six-week build, ~25–30% of one GTM engineer’s time. Lead-to-opportunity conversion maintained; touches-to-convert reduced (agent responds instantly; humans let leads sit in a queue). Agent cost: ~$1,000/year vs. >$1M in salaries.

The nine SDRs were moved to outbound — a higher-value function they were previously unable to staff.

Worked example — deal-bott and lost-bott

The lost-bott runs Gong call transcripts, Slack messages, and email threads for every closed-lost deal through an agent that produces a root-cause loss analysis. AE self-reporting (typically in CRM) is systematically biased; the agent frequently disagrees. A deal reported as “lost on price” was reclassified as “never engaged economic buyer; failed to demonstrate ROI.”

The real-time deal-bott pushes health alerts into per-customer Slack channels: “You are mid-process and have not spoken to an economic buyer.” At high product-shipping velocity, the deal-bott also identifies objection-handling gaps post-launch, feeding a weekly sprint to fix “bugs in the go-to-market process.”

Ideal profile

Former sales engineers are the natural first GTM engineers: they understand what good sales looks like (unlike pure engineers) and they can write code (unlike pure AEs). A hybrid “general athlete AE + GTM engineer” profile can also work at smaller companies: one person develops their best-practice playbook and then tries to automate it.

Where mainstream views differ

Some observers frame GTM engineering primarily as a threat to SDR and AE jobs. Grosser’s view is more nuanced: the tasks AI takes on are the ones humans found least interesting anyway, and the redirection accelerates junior salespeople into higher-value work sooner. The last part of GTM to be automated is complex multi-stakeholder enterprise prospecting where navigating an org chart requires human judgment.

The opposite concern — that AI-generated outreach will simply flood inboxes with low-quality emails — is addressed by the human-in-loop model: the agent proposes, the human approves, and quality remains high while throughput increases.