No. An agent that can pursue pipeline for weeks needs a score gate before it spends anything. Score every account first, let automation run cheap on low fit, and put human time and enrichment budget only on the top of the list.
I'm Michael Thomas Zajac, a sales operator in Hermosa Beach. I have built this gate by hand, and I think it matters more now that agents can stay on a job for weeks.
What changed on September 11?
On September 11, 2026, Salesforce expanded Agentforce with a portfolio of job-ready agents. One of them is Hunter, an outbound sales agent that "works a sales pipeline from research to outreach, collaborating with sellers over weeks and months." Hunter is in pilot now, with GA slated for November 2026. It is also the first agent on Salesforce's new long-horizon runtime, built so agents can pursue goals across days and weeks instead of finishing a single task.
Salesforce shared an early customer result too: 60% of Perk's sales pipeline is built by its outbound sales agent, Hunter.
That is a real shift. And to be fair, the announcement talks about guardrails that define when the agent can act on its own and when seller approval is required. Good. The operator question is what goes into those guardrails.
What breaks when an agent runs for weeks without a score?
The same thing that broke every ungated automation before it, only slower and for longer. A one-shot sequence can burn a list in an afternoon. A long-horizon agent with a goal and no fit score can keep spending for weeks: enrichment credits on accounts that will never buy, touches on buyers who were never a match, and brand trust with every message that lands on the wrong desk.
Duration multiplies whatever the agent is pointed at. Point it at your best-fit accounts and it compounds pipeline. Point it at everything and it compounds waste. The agent is not the problem. The missing gate is.
What did the score gate look like at SuperSend?
I was the first commercial hire at SuperSend. I sold, and I built. The GTM engine was an internal web app that served as the hub. Every day it pulled in 5 to 10k domains, each with a reason to approach. Every domain got an ICP score from 1 to 5, and the score decided what that domain was allowed to cost.
- 4s and 5s got human budget. Sales Nav connects, InMails, DMs, cold calls, sequencing, retargeting. Fives got 30 Sales Nav connects a day. Clay enrichment found the top 3 personas, and when it found a phone number, the phone went on the cadence.
- 3s got moderate, automated effort. Identified after a hand-raiser or a signal, then automated. No human in the loop.
- 1s and 2s stayed cheap. No Clay credits. Cold email, link traffic, and capture the hand raiser if one showed up.
Ingest shut off when human budgets were maxed. The engine never asked people to chase more than they could work well.
What did the gate produce?
10 net-new enterprise accounts and approximately $400K in net-new ARR in 2.5 months.
That came from putting humans only where the score said they belonged. Finding the needle in the haystack was the hard part. The score is what found it.
Which scores should earn human time and budget?
This is the design question for any long-horizon outbound agent. Not "should the agent run?" but "what is each score allowed to spend?" Here is my version:
- Top scores earn humans. Let the agent do the research and the prep. A seller owns the relationship and approves the touches that carry your brand.
- Middle scores earn automation after a signal. Wait for a hand-raise or a real trigger. Then let the agent work it.
- Low scores earn nothing expensive. No enrichment spend. Cheap channels only, and watch for the hand raiser.
- Capacity caps the queue. When the humans are full, the agent stops feeding them. It does not lower the bar to stay busy.
The rule underneath all of this is in Humans in the loop, or you are just scaling noise.
What should you do Monday?
Before you pilot any agent that works pipeline for weeks, write down your score tiers and what each tier is allowed to spend: human hours, enrichment credits, channels. Then make that the agent's guardrail, not an afterthought. If an account has no score, the agent does not touch it. Keep the score. Keep the gate. Let the agent stretch the team, not the budget.