Everyone in go-to-market is building agents right now. Almost nobody is talking about who maintains them at 2am when the model provider ships an update and half your workflows quietly break.
That gap, between the excitement of building and the reality of owning, was the thread running through our recent webinar on building the agent-native revenue organization. We brought together three people who see this from very different seats: Chris Reisig, a five-time CRO and GTM advisor to high-growth companies; Cliff Simon, CEO and founder of Polaris Ops and host of the Lodestar Podcast; and Devang Agrawal, CTO and co-founder of Airspeed. Tamar Martirosyan, our VP of Growth Marketing, hosted.
The conversation kept circling back to one uncomfortable truth. Standing up an agent is the easy part. The question that actually decides whether your revenue org wins or drowns is: who owns this thing over time, and what does it cost you to keep it alive? Here’s how the three of them worked through it.
Agents aren’t a tool you buy, they’re a teammate you onboard
Start with the mental model, because the wrong one sends you down an expensive path.
For most of us, the revenue org has always been a specific cast of humans. Reps, SDRs and BDRs, customer success, ops. Chris’s argument is that the roster just changed.
Your team isn’t just made up of humans anymore. You had your sales reps, your BDRs and SDRs, your customer success team, your ops team. Those were the players in the orchestra that helped you deliver revenue. Now you have to embrace agents as another member of that team.
Chris Reisig, five-time CRO and GTM advisor
The word that matters there is orchestrating. This isn’t humans versus agents, and it isn’t a race to replace headcount. It’s about using agents to make the humans dramatically more effective, especially on the grunt work that eats their week.
In sales, the human touch will never be replaced. At the end of the day, people buy from people. There needs to be a relationship, a trusted partnership, before a transaction can occur. But these agents can help the reps, ops, CS, and the lead gen function get a lot of the grunt work in their lives done, so they can focus on making the human experience even more incredible for the customer.
Chris Reisig
And if agents are teammates, they need onboarding like teammates. Chris made the point that sales enablement now has a whole new dimension. You already teach new hires your product, your competitive landscape, and how your company operates, which is hard enough when you are trying to ramp reps faster. Now you also have to teach them your suite of agents: which ones you use, how to use them, and how those agents inform their daily work. Enablement isn’t just product and process anymore. It’s, here’s your agentic AI stack and here’s how you work alongside it.
You can’t skip the foundations, and most companies want to
Before anyone gets to deploy a single agent, Cliff had a warning that landed hard: a lot of teams have coasted on great product-market fit and never built the underlying structure. That doesn’t fly anymore.
His analogy was the best moment of the session.
When my son was five, we started teaching him how to unload the dishwasher. He can’t know where the forks, knives, and spoons go unless he knows what a fork, a knife, or a spoon actually is. AI is very similar. AI is a super-intelligent six-year-old. It can do amazing tasks at scale, but only the things we’ve given it a very concrete understanding of. Grab A, put it in B. You need to give it those guardrails, and as an industry, we’re really bad at that.
Cliff Simon, CEO and founder, Polaris Ops
Why are we bad at it? Because the IP that agents need to work from, the process, the standard operating procedure, the actual meaning of your data, lives in people’s heads. When it does get written down, it’s unstructured, and there are twenty different versions floating around between the BDRs, the AEs, marketing, sales leadership, CS, and probably a fourth opinion from finance.
We need to have one homeostatic version of that, that everyone can run off of. You talk about the context components, the knowledge graph, the governance, but you don’t even get to those pieces unless you have something the agents can actually work from.
Cliff Simon
That’s the foundational move: a taxonomy, one source of truth, a shared understanding of what everything means. In practice that starts with making your CRM the source of truth and running a real CRM hygiene checklist before you point a single agent at the data. Guardrails first. Ambition second.
From insights to action, and from hype to reality
Devang is, by his own description, deeply skeptical of hype. Which makes it more interesting that he thinks something genuinely changed.
I try to stay off the hype bandwagons. For the longest time I felt agents were not the right technology, because you couldn’t trust an LLM to take customer-facing actions or actually do something in your systems. Only in the last few quarters has there been a complete step change, because of the quality of the underlying LLMs and how they’ve been specialized for agents. I felt agents were just hype until 2025. In 2026 they’re really becoming a reality.
Devang Agrawal, co-founder and CTO, Airspeed
He went further on the pace of it. We’ve all lived through technology transformations, but this one feels different in kind.
No transformation has ever moved faster than the agent-native transformation. Already, every single role I’m hiring for, across GTM, finance, and engineering, people need to be fluent in building and managing agents. That shift happened over the last six months. If you’re not doing that, you’re massively falling behind.
Devang Agrawal
The part that used to feel like a dream, and now doesn’t, is agents moving from giving you insights to actually reasoning over messy, high-stakes questions.
An agent can tell you why you’re losing deals. Why is your EMEA team losing deals? Why is your mid-market team losing deals? It can take a really unstructured piece of information and roll with it.
Devang Agrawal
Inside Airspeed, that looks like agents mining every data source to score channel effectiveness, flag deal risk before a deal closes, and help leadership deploy resources so no deal ever quietly slips. That is also what turns a forecast into something you can defend. The shift from “here’s a summary” to “here’s what to do about it” is the whole game.
The build-versus-buy trap nobody warns you about
Here’s where the conversation got refreshingly honest, because all three have watched the same movie play out.
A year ago, the forward-leaning move was to grab an LLM and build your own thing. And it worked, at first. Then reality arrived.
That worked well in the beginning, but teams quickly figured out, now I have to maintain it. Is there a regulatory requirement in my industry I have to meet? The sales team likes it so much they want me to add this and add that. Suddenly your best ops person, who you rely on for a hundred things in the business, becomes a software developer. I’m not sure that’s what you want.
Chris Reisig
And then there’s the decision underneath the decision, the one no revenue leader should be spending their energy on.
Imagine having to pick the best web browser twenty or thirty years ago. You might have picked the wrong one. You don’t want your ops person or your CRO deciding which LLM to use.
Chris Reisig
This is exactly where a harness earns its keep. It uses multiple LLMs and picks the best result, manages the cost on the back end, and handles the compliance and regulatory oversight, so your team doesn’t have to own any of it. We have written up the whole calculation in our build versus buy breakdown, the honest case for buying your AI sales infrastructure, and a line-by-line look at what a DIY AI sales stack really costs to run.
Cliff added the guardrail every builder needs to hear. Agents are brilliant at crunching things together, but they don’t bring critical judgment, and you cannot skip the human in the loop.
I just finished a white-space report for a customer. Multiple product lines, hundreds of accounts. The numbers the AI put together, I would never put in front of a board. It wanted one customer’s ARR to go from five million to twenty-five million, when that account was already at max capacity. That’s not realistic. Having a person be the layer that checks things is really important.
Cliff Simon
He’s also worried about a generation trained to operate systems and tools but not to think critically, which puts the onus on GTM leaders to nurture that judgment. And his rule for the agents themselves is worth pinning to the wall.
I don’t want my agent going out and creating a manifesto. I don’t want these agents thinking for themselves. I want them to have a very set protocol and only do the things they’re supposed to do.
Cliff Simon
Agent builder burnout is real, and it’s already here
This was the part of the session that got the most nods, because it names something a lot of teams are quietly living through.
We’re now starting to see the first few cases of agent builder burnout. Someone I know, one of the most ‘let’s build agents ourselves’ people I’ve met, is finally having to climb out of it and tell people, guys, I’m actually not a software engineer. I also have a full-time job. I’m not going to constantly take feature requests and maintain this stuff I’ve built.
Devang Agrawal
When you buy from a vendor, you send a feature request and expect it to get done. When it’s your own internal build, the person who made it usually has no capacity left to service anyone. Devang has watched CROs fall into the same trap, building genuinely impressive agents on nights and weekends with Claude Code, and then facing the question of who maintains a substantial agent library while also, you know, running a large revenue team.
Chris summed up the whole dynamic in six words.
It becomes a beast that needs to be fed.
Chris Reisig
And then Devang raised the risk that should genuinely scare you: the bus factor.
If one person built all of this in a code base no one else can access, because you weren’t operating as an engineering team at scale, and then that cracked RevOps operator leaves, what are you going to do? You have all these agents your team loves, and no way to actually do anything with them anymore.
Devang Agrawal
That is the same failure we wrote about in why agent harnesses built for individuals break at enterprise scale: tools built for one person give leaders a black box instead of an organizational capability.
Cliff put a number on the ongoing tax, and it’s higher than most CROs assume.
A good go-to-market engineering team spends a third of their time making sure everything they built still works. Anthropic, OpenAI, Perplexity, whatever they’re doing to the models, we have to go make sure everything still works based on their changes. A SaaS company would have taken care of that. It’s in the T’s and C’s. CROs are used to buying something, setting it up in ninety days, and then set it and forget it. This is wildly different. Implementation’s never done, because it’s always learning.
Cliff Simon
That’s the reframe. The cost of an agent isn’t the cost of building it. It’s the total cost of ownership, forever, in a world where the ground moves under you every few months.
The answer: an agent harness, and a maturity curve to get there
So if building everything yourself leads to burnout and a bus-factor time bomb, what’s the alternative? This is where Devang laid out the concept at the heart of what we’re building at Airspeed.
Not every organization is ready for agents on day one, so the honest version of this isn’t “deploy agents.” It’s a journey, mapped by a GTM agent maturity model. Most teams start with call recording and historical data, often for compliance. From there they move into conversational intelligence, tracking keywords, sentiment, and the basics of coaching. Then into revenue execution, where pre-built agents start helping you run the motion, MEDDIC or MEDDPICC analysis and the like, though still somewhat siloed.
The goal state is the agent harness.
An agent harness is actually a simple concept. It’s a layer that orchestrates all of your tools. Instead of having an agent living in every platform, you have one harness orchestrating all of them, getting data from everywhere into one place, and then executing across the board. It’s where all the intelligence for your organization lives, rather than being managed in every separate tool.
Devang Agrawal
The reason the harness is the destination and not just another feature is what it unlocks.
Once you get to that place, you’re creating compounding loops that maximize your revenue per head. You orchestrate all your other tools and keep the intelligence in one place, rather than managing intelligence in every other place.
Devang Agrawal
That last distinction is the whole point of the session in a sentence. A collection of AI features scattered across your stack is not the same thing as one harness that ties them together. One creates maintenance debt in a dozen places. The other compounds. For the longer argument, and what a GTM-native harness needs that a general-purpose one doesn’t, read why GTM needs a native agent harness and what a commercial brain is.
What the near future actually looks like
Looking ahead, Chris sees agentic strategy becoming a talent magnet, not just an efficiency play. He’s interviewing sales candidates every week, and the questions have flipped.
The best sellers understand the power of agentic AI to make them more productive, and they want to maximize their production so they can maximize their comp. A stated strategy around agentic AI is becoming a precursor to even wanting to work for a company. If you don’t have it, my competitor will, and I’ll lose business. It’s not just about having the best product anymore.
Chris Reisig
Cliff was refreshingly humble about prediction, and honest about the human cost of the pace.
We are terrible at predicting the acceleration of change. If you’d asked me in January whether we’d be doing things the way we are now, I’d have said we were two, three, four years away. Those three-to-five-year cycles are compressing into three-to-six-month cycles. It feels like we’re all wartime all the time, and you can’t keep that up forever.
Cliff Simon
His hope, and Devang’s strong agreement, is that the context layer is where the next real progress comes from. Summarizing a single call is trivial. Answering a question about a whole person, a geo, or a product line, and giving the model exactly the right context out of everything you know, is the hard and valuable part. That, again, is what a harness is built to do.
And on the very practical question of getting reps to actually adopt any of this, Devang was direct.
If you build agents as a standalone product, giving output in a surface reps aren’t already in, we don’t see adoption. But if you deeply embed it in the tools they use every day, Slack, iMessage, WhatsApp, and in the SaaS software they already live in, that’s when usage increases. Reps don’t want to change their workflows and think about tooling. They want to sell, and the agent just has to be in the background helping them.
Devang Agrawal
That is why Airspeed runs inside Slack and connects to the systems reps already live in, rather than asking them to log into one more tab.
The takeaway
Put it all together and the story of the agent-native revenue org isn’t really about whether to use agents. That debate is over. It’s about ownership. Build the foundations before the ambitions. Keep a human on the critical judgment. And be brutally honest about the total cost of owning agents yourself, because the beast needs feeding whether or not you planned for it.
The teams that win won’t be the ones with the most agents. They’ll be the ones with a harness, so their intelligence compounds in one place instead of decaying in twelve.
That’s exactly what we’re building at Airspeed: an agent harness paired with a revenue execution platform, designed to walk organizations down the entire maturity curve, from getting their data foundations right to orchestrating agents across the whole revenue motion. If you would rather not build the first agents yourself, our forward-deployed team will build them with you, and the agents worth building first are a good place to start. Want to see where your team sits on the curve? Book a demo.
