The news
Martech's Pamela Parker lays out the economics of agentic AI workflows in plain terms: a single daily pipeline — search, summarize, generate — can blow through a $20/month subscription before the month is half over. The culprit is tool calling, which forces AI agents to pass the full task history, reasoning chain, and external tool data back through the model at every step. Providers built pricing for chatbots; teams are deploying agents.
Our take
This is the infrastructure conversation that GTM teams haven't had yet — and it's coming for them fast.
Most marketing and RevOps teams pricing out AI tooling are still thinking in chatbot terms: a flat monthly seat, maybe a usage tier, maybe not. That math breaks the moment you move from "ask a question, get an answer" to "run a workflow that touches HubSpot, searches the web, and outputs a report." Agentic loops are multiplicative by design. Every tool call, every reasoning step, every external data pull gets re-injected into the model's context window. The token meter doesn't pause while the agent thinks.
The critical point in the article — that the fix is where your data lives, not how many tools you use — is the one most teams will skip over. The instinct is to pull back: fewer tool calls, simpler workflows, smaller ambitions. That's the wrong move. The teams who will win here are the ones who invest in owned context — structured, queryable data stores that agents can access without re-reading everything from scratch on every loop. That means cleaning up your CRM data model, thinking seriously about how campaign data is stored and surfaced, and treating your internal knowledge as infrastructure rather than an afterthought.
This is also where no-code and vibe-coded tooling matters. If your team is running agentic workflows through off-the-shelf chat interfaces, you have no visibility into what's getting passed back to the model on every step — and no lever to optimize it. Building even lightweight custom tooling gives you that visibility.
The deeper issue: most GTM teams are not yet thinking about context engineering as a cost-control mechanism. They will be.
The so-what
The token bill is your first signal that an agentic workflow is worth operationalizing — it means something is actually running. Here's how to get ahead of it:
- Audit before you scale. Before connecting your next tool via MCP or API, ask: does the agent need all of this data, or just a summary? Smaller context = fewer tokens = lower cost.
- Invest in your data layer now. Agents are only as cheap as your underlying data is clean and structured. Messy CRM fields and unorganized campaign data get expensive fast when an agent has to sort through them on every loop.
- Track token consumption as a workflow KPI. If you're not measuring it, you're not managing it — and your next bill will surprise you.
The teams who treat context engineering as a GTM capability — not just an AI-team problem — are the ones who will actually be able to afford to run agents at scale.