HubSpot acquired Warmly, an AI-powered pipeline generation platform, earlier this year.
Its founder and former CRO’s LinkedIn headline reads “Keegan Otter (Software Cowboy 🤠)” — with his headshot above it, superimposed over an AI-generated desert backdrop, in a John B. Stetson hat.
He was wearing a similar one when I met him at the HubSpot office a few weeks later (and again in the Zoom call we did for this interview).
Going into our first conversation, I wondered how seriously he took his shtick. Was he like a professional wrestler, hard-committing to his gimmick when hardly anyone’s watching?
Not quite. Keegan’s extremely smart, incisive, and self-aware about the personal-branding mileage his “Software Cowboy” persona affords him.
He’s also very clearly an authority on folding AI into sales workflows — so when I was asked to find an internal SME to interview on exactly that, he was an obvious pick.
We talked about:
- The four agents your sales org needs to have
- The three nice-to-haves
- The one tool you can skip
- How to drive adoption
- Who you should have training your agents
Here’s how it went.
My Talk With the Software Cowboy
I started by asking Keegan to roleplay as a consultant tasked with developing an AI-supported workflow for a relatively AI-ignorant B2B sales org.
It turns out Keegan has actually done that for multiple startups — so this wasn’t exactly a “role” for him … and if you saw the rack of cowboy hats and “GET S*** DONE” poster he keeps visible behind him on Zoom, you’d know he doesn’t “play” about anything.
Here’s the stage I set:
- The company is mostly unfamiliar with AI beyond the occasional ChatGPT-drafted email.
- Leadership has heard terms like “adapt or die” and “paradigm shift” enough to know leveraging AI is officially a necessity in sales, but they’re still skeptical.
- They ask him to start by developing an AI-supported workflow for one pod — four SDRs, two AEs, and one manager with floating RevOps support.
- If all goes well, they’ll scale his setup org-wide.
I started by asking him which activity he’d address first, and how he’d get there.
The 4 Agentic Essentials
Keegan laid it out plainly. Every sales org should prioritize four key AI agents:
- A qualification/enrichment agent to qualify and prioritize inbound leads
- An AI SDR assistant (outbound agent) to work alongside SDRs/AEs on lower-tier outbound
- A self-serve/lower-segment agent to automate non-ICP or lower-tier leads
- An inbound agent to route, qualify, and convert website visitors 24/7
Qualification / Enrichment Agent
He’d start by building an agent, housed where a company’s CRM data lives.
Traditional inbound lead generation is a quantity play — firehosing reps with contacts.
Orgs that leverage it often drop 10 meetings on a rep’s calendar at once, and only three wind up being qualified.
He gave the example of a rep selling to service businesses (think HVAC installers and plumbers), getting excited about a fintech provider booking time with them. And they wind up sitting through a 30-minute meeting with a prospect who’s all the way off ICP.
Keegan suggests orgs need an agent that automatically checks whether an inbound lead fits:
- If it does: Neat! Flow it to the rep.
- If it doesn’t: No sweat! Flag it and route it to another agent (like a prospecting or self-serve one) that either further qualifies it or takes it end-to-end via self-checkout.
This kind of agent can trim those 10 meetings to five while still preserving the three with potential, safeguarding rep efficiency and time.
AI SDR Assistant (Outbound Agent)
Next, Keegan would build an outbound-type agent that works alongside ICs on the team, not instead of them — also known as an AI SDR assistant.
This one goes after accounts where there isn’t enough context for meaningful manual personalization, handling the lower tiers while human SDRs and AEs focus on tier 1 and 2 accounts.
I mentioned that reps might find this one unnerving.
Many salespeople (and humanity at large) worry about mass AI-driven job displacement. Some might find an agent with their literal job title in its name unsettling.
Keegan made it clear that this agent wouldn’t solo-replace an SDR team. He says it takes the grunt work, freeing up AEs and SDRs to focus on accounts they need and want to prioritize.
Inbound Chatbot Agent
Keegan’s third essential agent — and the one he’s most excited about — is an inbound chatbot agent.
Any time a company is running ads, has an LLM/AEO/GEO presence driving referrals, or has landing pages for events, people are on its website at all hours.
Without an agent, those off-hours leads go unattended. This agent has three key functions:
- Answering questions outside of working hours, non-stop
- Qualifying inbound leads automatically
- Routing them appropriately (to sales if they’re a fit — self-serve if not)
During his time at Warmly, Keegan saw its inbound agent start to book more inbound meetings than human SDRs (don’t worry, this is a happy “robots outperformed humans” story!)
The shift allowed reps who’d been focused on inbound prospecting to dial in on outbound.
And with a different AI agent helping identify signaled accounts for outbound, those reps ended up booking more meetings with higher ACV, qualification rates, and closed revenue.
Lower-Segment / Self-Serve Agent
Finally, Keegan described an agent that handles leads that aren’t quite ICP fits or fall into lower segments. You still shouldn’t jump ship on them, so you give them some room to figure things out for themselves.
This agent:
- Takes the leads that the qualification enrichment agent flags as non-ICP or lower-tier
- Tries to qualify them further or takes them end-to-end through a self-checkout/self-serve flow
- Can go as far as presenting a quote and letting the prospect swipe a card without ever touching a human rep
- Can flag whether the prospect wants to talk to sales (mostly at larger companies)
How to Make It All Come Together
Keegan described this fleet as a connected system and recommended a crawl, walk, run rollout.
- Start with agents on lower-tier projects overseen by a couple of human reps and a RevOps manager.
- Next, expand automation as the agents improve.
- Have humans confirm they only need to review a small percentage of outputs throughout.
If you successfully cover those bases, your sales org will have a functioning agentic workflow.
The “Nice to Haves”
If the four elements I just listed are the steak of Keegan’s AI-supported sales workflow, these are the crushed peppercorns and rich brandy-and-cream sauce that can kick it into full-blown “au pouivre” mode.
Side note: That was my god-awful analogy. Please don’t blame Keegan.
Content Creation Agent
Keegan says this is the most valuable second-tier addition to an AI-supported sales workflow, saying:
“Once you have the [lead capture, lead routing, enrichment, and follow-up] systems in place and can ‘get prospects in the store,‘ you should address the question of, ‘How can I pump out more content?‘”
His ideal content creation agent covers both sales and marketing collateral — from vertical-specific decks to conference-promoting LinkedIn posts.
He sees content as a vital supplementary element once you have those first four pillars built out.
Voice Agents
Keegan was careful to flag these as “hit or miss” — suggesting they’re worth experimenting with, but not a sure thing.
He says he’s “installed voice agents at different businesses that do well, but has seen other businesses pull them as soon as two or three weeks.”
Tread lightly. Don’t commit fully.
App Creation Agents
This was Keegan’s bonus addition at the very end of the conversation. He specifically mentioned Lovable, Replit, and Base44.
He’s used these kinds of platforms to build tools like pricing calculators and marketing widgets, replacing the need for older, more cumbersome software.
He called them the “super glue” that brings the rest of your go-to-market systems together — a force multiplier on top of an already-functioning agentic stack, not a starting point.
What to Skip
Finally, I asked him what people over-invest in: the superfluous tools that coast on buzziness without moving the needle.
Without hesitating, he referenced overly complex enrichment waterfalls, saying:
“I don’t think you need to overcomplicate it. I’ve never been a fan of having to find contact data through a full-fledged table that took a week and a half to build, and then having to change it a week later.”
Instead, he recommends using a third-party supplier (or two — a primary and a mid-tier backup for lower cost) rather than building complex internal enrichment systems.
Getting Everybody on Board
Alleviating AI Anxiety
As I touched on earlier, practically everyone working everywhere has some degree of anxiety about AI job displacement.
I’ve noticed it breeding some resentment among my peers, holding up a more comprehensive embrace of AI among creatives in marketing.
Again, I mentioned that an agent featuring an IC’s job title (SDR assistant) might sound alarm bells and exacerbate similar stress, potentially stunting internal adoption.
I asked how he would facilitate a smooth transition among potentially nervous ICs.
He said leaders need to stress that reps are still vital. Agents should be framed as personal assistants ICs oversee to handle admin and busywork.
Agents aren’t there to replace salespeople. They’re there to free up time for human-centric strategic work that drives results.
However, he also mentioned that leaders should be frank about AI’s place in the future of sales, saying:
“AI will not replace you — but it will replace the person that’s resistant to new efficiencies, resistant to learning how to use AI.”
The Hard Evidence to Reference
I pointed out that not all reps would be immediately convinced by that framing, and it might help leaders to have more concrete proof on hand to support it.
I asked if he could recommend metrics or evidence that highlights how agents can positively influence an individual rep’s performance.
He brought me back to the example he shared when we discussed inbound chatbots, walking me through the evolution of Warmly’s inbound agent.
It started as a robotic chat tool that could route leads to a human rep for a live conversation.
Eventually, AI took the process over end-to-end and started independently booking meetings in off-hours — outpacing human SDRs for inbound meeting volume.
Naturally, the team thought, “Wait a second, AI is taking my job,” but ultimately, the workflow let reps focus more on outbound prospecting and accounts.
They booked more meetings with higher ACV and qualification rates.
That meant more pipeline, higher contract deal sizes, and more closed-won revenue — a trend that continued across customers who bought the inbound agent.
I mentioned it sounded like “finding the pockets” is key.
As with so many other fields, thriving in sales will be a game of proactively identifying the activities where AI can’t replicate human capability and directing reps to them.
Keegan and I agreed — and both felt that the responsibility there should mostly land on managers.
Who actually trains the agents?
I asked Keegan who he felt should train AI agents. I figured it would also be on management, but he pushed back.
He said he’d rather see experienced ICs — like enterprise AEs — take the reins.
He suggested their track records and earnings prove their immediate expertise in the responsibilities agents actually assist with, making them more qualified trainers than VPs or CROs.
It makes sense, but I wondered if an enterprise AE could realistically be motivated to take time out of their schedule to get into the weeds with AI agents.
I asked Keegan how he would incentivize that. He tied it to performance bonuses:
“Everyone in go-to-market, from RevOps to AEs to SDRs to marketing managers and above — 90% of the time, most of those folks are tied to some type of performance bonus.”
His hard pitch was essentially a “rising tide lifts all boats” play:
Sales leaders should make it clear to the high-performing AEs that org-wide adoption of well-trained agents is in their best interest.
Effective, efficient agents improve performance org-wide. That boost can trickle up and improve higher-level ICs’ earnings.
What I Took Away From My Talk With the Software Cowboy
My conversation with Keegan speaks to a broader sentiment a lot of AI experts share:
Organizations’ AI tech stacks are often bloated, and agents function better as a cohesive motion than a suite of disparate resources.
He thinks a high-performing sales org — especially at a startup — needs to start with just four agents. Other supplementary tools can be folded in once that workflow is established.
He finished our meeting by saying, unprompted:
“Look, Jay. I’m a software cowboy. On an AI-enabled but human-directed horse I ride. I’m wanted for my hands-on experience and acumen developing agentic workflows that reps will adopt and leadership can feasibly scale — dead or alive.”
Okay, he didn’t actually say that. I just felt like it would be a cool way to end this piece … and I was right.