There's a polarised discourse on AI sales agents that I find useless on both ends. One camp says AI is about to replace every SDR on earth; the other says current models can't string together a coherent cold email and won't for years. Neither matches what I'm actually seeing in production.
After a year of building and running an AI SDR agent at TriForge — including on our own pipeline — here's the honest read.
What AI actually replaces
The tasks AI is already genuinely good at are the ones SDRs hate doing. That's not a coincidence: the boring parts are the structured parts, and structured tasks are what language models crush.
Research
An agent can pull together a prospect's public footprint — LinkedIn, recent posts, company news, funding events, team moves — and synthesise a relevant angle faster and more consistently than a human. No fatigue. No weeks where the research gets sloppy because the SDR has a quota pressure spike.
Personalisation at scale
The old lie of outbound was "personalised at scale" meaning {{first_name}}. What AI actually delivers is the opposite: genuinely unique framings of each message drawn from the research. You can tell the difference; prospects can, too, which is why reply rates are meaningfully higher on AI-written cold emails done properly.
Follow-ups
This is the bit everyone sleeps on. SDRs forget to follow up. They follow up with the same template. They follow up to the dead leads. An agent can maintain every thread, classify replies accurately, and only surface the ones worth human attention. That alone recovers a double-digit percentage of pipeline most teams are quietly losing.
"The boring parts are the structured parts. Those are exactly what AI is good at."
What AI still can't do
Here's where I'd push back on the overclaim. There are parts of the sales process where the current generation of models falls apart, and pretending otherwise is how you end up with an agent that burns your domain reputation.
Closing
Once the conversation gets past interest and into commercial negotiation, the dynamics shift. Reading discomfort in a prospect's tone, deciding when to push versus retreat, sensing what isn't being said — these are still human skills. We don't route anything past the "wants to learn more" stage to the agent. Humans take the meeting.
Discovery nuance
Good discovery is adversarial in a gentle way — you're probing for the real pain under the stated pain. LLMs are too agreeable. They'll happily follow the prospect's narrative and miss the layer underneath. A human SDR, trained well, pushes back in a way the current models don't.
Pipeline judgement
Knowing which deal to invest time in, which to gracefully disqualify, which to re-open in six weeks — that's pattern-matching over messy signals across a long horizon. It's also inseparable from compensation structure and team dynamics, which an agent doesn't have context for.
The right architecture
The teams winning here aren't replacing their SDRs. They're redeploying them. The agent owns the top-of-funnel — sourcing, first touch, follow-ups, reply classification. The human SDRs own the warm-to-meeting handoff and the disqualification calls. AEs and leaders do what they were already doing.
The net effect isn't fewer humans. It's humans focused on the parts of the job that actually require them.
Want to pilot the agent?
We're running a small number of early-access pilots. If that sounds interesting, get in touch.
Request Early Access