Issue #9: Enterprise Agent ROI Proves Out

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

The Hook

Salesforce just hit $800M ARR on Agentforce — 169% YoY growth, 3.2 trillion tokens processed in a single quarter. The enterprise isn’t evaluating agents anymore. It’s renewing contracts.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

This Week’s Signal

Agentforce + Data Cloud combined ARR is nearly $1.4B. That number matters because it’s not a demo stat — it’s recurring revenue from production deployments. But the more interesting signal is the token count: 3.2 trillion in Q1 alone. That’s the volume of actual agent work happening inside real enterprise workflows. For operators, the implication is structural: the companies scaling agents fastest are the ones that already had clean data infrastructure underneath. Salesforce’s Data Cloud is the unlock — agents without a live, queryable data layer are just expensive autocomplete. Build the data foundation first. Then deploy the agents on top.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

3 Operator Playbooks

1. Measure Containment Rate Before You Measure Cost Savings

Wonderful raised $150M at a $2B valuation on one core metric: 80%+ containment rate across 30+ enterprise markets, with handling times cut by 60%. Containment rate — the percentage of interactions an agent resolves without escalating to a human — is the number that determines whether your deployment is a product or a prototype. Your move: Before you optimize for cost, benchmark your current containment rate. Even a simple support agent should target 60% containment in week one. If you’re below that, the problem is almost always missing context, not model quality. Fix the data layer, not the prompt.

2. Put Agents Where Your Team Already Lives

Gumloop closed $50M from Benchmark on one insight: non-technical employees won’t use a new tool. They’ll use the tools they already have. Their platform lets anyone build AI agents inside Slack, Teams, and email — no code required. Customers include Shopify, Ramp, and Instacart. Your move: Stop building internal agent portals nobody opens. Deploy your next automation directly inside your team’s existing workflow surface. If your ops team lives in Slack, the agent should be in Slack. Reduce the adoption gap to zero and adoption becomes the default, not the obstacle.

3. Watch the MCP Ecosystem Mature Into Transactional Infrastructure

FreeWheel — Comcast’s ad-tech arm — just became the first major player to embed an MCP server directly into its transactional layer for premium video ad deals. AI agents (starting with PMG’s ‘Alli’) can now adjust campaigns in real time against live inventory data. This isn’t an integration — it’s infrastructure. Your move: If you’re building on any platform that runs live transactions (ad buying, inventory management, pricing, scheduling), ask your vendor if they have an MCP server or a published MCP spec. The ones that do will let you connect agents in hours. The ones that don’t will cost you quarters.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Steal This

Containment Rate Audit — a 20-minute workflow for any agent deployment:

1. Pull the last 100 agent interactions from your logs
2. Tag each: Resolved (agent handled end-to-end) / Escalated (human took over) / Abandoned
3. Calculate: Resolved ÷ Total = Containment Rate
4. For every Escalated interaction, log the reason in one line (missing info, wrong response, unclear intent)
5. Sort by frequency — your top 3 escalation reasons are your next 3 improvement targets
6. Fix one per sprint. Rerun the audit monthly.

Wonderful’s 80% containment didn’t come from a better model. It came from iterating on the failure cases. This audit is how you find them.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Also on Our Radar

• Tencent launched QClaw (one-click PC remote control via WeChat) and WorkBuddy (enterprise agent) — both tested internally by 2,000+ employees. Pony Ma announced a full agent product matrix targeting 1.3B+ WeChat users. The distribution moat in consumer AI just got a lot harder to compete with.

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

If this issue was useful, forward it to one operator who’s still debating whether to deploy agents. The debate is over. Help them catch up. → insider.dforge.ca

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ AI Agent Insider is a weekly briefing for founders and operators who build with AI agents. No hype. No sponsored picks. Just signal.