The news
QuantumBlack, McKinsey's AI arm, published field notes on agentic workflows in software development — covering what's actually working after two-plus years of teams moving from AI copilots to autonomous agents. The piece is practitioner-level and pulls no punches: adoption is real, but so are the failure modes.
Our take
The software world is about two years ahead of GTM on agentic adoption. That gap is a gift — if you pay attention to what's breaking over there before it breaks over here.
The pattern QuantumBlack describes is familiar to anyone who's tried to automate something in a GTM context: teams reach for agents before they've done the boring work of defining what "good" looks like. In software, that means agents writing code nobody reviews until it's in production. In GTM, that means agents writing follow-up sequences, updating Salesforce records, or qualifying leads based on logic that was never written down in the first place.
The article is specifically critical of teams that treat agentic workflows as a drop-in replacement for human judgment — and that criticism hits differently when you apply it to revenue-critical processes. A misrouted lead or a botched nurture sequence doesn't break a build. It burns a relationship.
What the McKinsey team gets right: the teams seeing real returns aren't the ones who gave agents the most autonomy. They're the ones who drew hard lines around what agents can touch, built reliable review checkpoints, and instrumented everything so they could see when something drifted. That's not a technology pattern. That's an ops pattern.
This exact dynamic plays out constantly in GTM automation. The first question is never "which agent tool should we use?" It's "what does this process actually look like when a human does it correctly?" If you can't answer that, an agent won't help you — it'll just make your bad process faster.
So now what?
Before your team starts evaluating agentic tools for GTM, run this three-step gut check:
- Document the human version first. Pick one workflow you want to automate — lead routing, meeting follow-up, whatever — and write down every decision a good rep or marketer makes when doing it manually. That doc becomes your agent's instruction set.
- Define your review gates. Decide upfront which outputs the agent ships autonomously and which ones a human sees before they go out. Start with more gates, not fewer.
- Instrument before you scale. Log what the agent does. Not because you expect failure — because you need to know what "normal" looks like before you can spot drift.
The teams who nail agentic GTM won't be the ones who moved fastest. They'll be the ones who built the operational scaffolding first.