Fireside Chat with Tim Sanders & Dave Boyce — From LLMs to Agents: Architecting AI for Judgment, Velocity, and Trust

251
00:29:01.530 –> 00:29:20.009
Julia Nimchinski: And we are transitioning to our next 4SR chat. We are for a real treat here. Welcome, Tim Sanders, VP of Research Insights at G2, Executive Fellow, Harvard Business School, and Dave Boyce, author of Freemium. Congrats, and board member of Winning by Design. How are you doing?

252
00:29:20.280 –> 00:29:21.330
Dave Boyce: Amazing.

253
00:29:21.700 –> 00:29:22.370
Dave Boyce: How you doing.

254
00:29:22.370 –> 00:29:25.300
Tim Sanders: Great. Doing great, Dave, nice to see you.

255
00:29:25.470 –> 00:29:27.860
Dave Boyce: Yeah. Where are you sitting right now?

256
00:29:28.270 –> 00:29:29.930
Tim Sanders: I am in Dallas, Texas.

257
00:29:29.930 –> 00:29:31.529
Dave Boyce: There you go, rock on.

258
00:29:31.650 –> 00:29:35.240
Dave Boyce: Dallas, Texas. I grew up in Oklahoma, so we can be enemies.

259
00:29:35.680 –> 00:29:36.510
Tim Sanders: Okay.

260
00:29:37.000 –> 00:29:37.810
Tim Sanders: Alright!

261
00:29:37.810 –> 00:29:48.209
Dave Boyce: But also… but I think we’re so far beyond that, we don’t have to be enemies. I spent some time at Harvard so we can be friends, there you go.

262
00:29:48.210 –> 00:29:49.289
Tim Sanders: There you go.

263
00:29:51.880 –> 00:30:01.759
Dave Boyce: Awesome. Hey, Tim, how do you… like, I’ve got a ton of things that I want to ask you. I don’t know how you want to structure this, we could just be pretty organic, but .

264
00:30:01.760 –> 00:30:07.809
Tim Sanders: Yeah. I wish we had spent more time together. I was, like, doing homework on you, and yeah.

265
00:30:08.020 –> 00:30:13.170
Dave Boyce: This is a friend! Like, this is a friend! I should know this guy way better!

266
00:30:13.170 –> 00:30:22.770
Tim Sanders: Well, we’ll get to know each other better. The only thing I’d say as a prompt is that what you might not know, because I haven’t talked about it yet publicly, I’ve been doing research for 3 months.

267
00:30:22.770 –> 00:30:36.639
Tim Sanders: for a report that’s coming out in October at Dreamforce, and it’s on agents. We did a massive study of B2B software buyers. We’ve also aggregated thousands of G2 reviews just on agent categories, and we’re really getting a glimpse

268
00:30:36.640 –> 00:30:43.179
Tim Sanders: Into the adoption rate of agent, what results are really rolling in, and most importantly, what trust levels

269
00:30:43.180 –> 00:30:47.099
Tim Sanders: look like today and tomorrow, so there’s some fascinating insights there.

270
00:30:47.560 –> 00:30:49.800
Dave Boyce: That is a good prompt.

271
00:30:51.930 –> 00:30:53.060
Tim Sanders: And that…

272
00:30:54.010 –> 00:30:56.219
Dave Boyce: Is that a G2 presentation? Is that a sales.

273
00:30:56.220 –> 00:31:00.980
Tim Sanders: Yeah, it’s a G2 report, and it’s called Leap of Trust.

274
00:31:01.200 –> 00:31:14.089
Tim Sanders: agents are winning, hearts and wallets in 2025, and the TLDR is that it’s expanding faster than anybody ever dreamed, and 1 out of three large enterprises are letting it rip in their workflows already.

275
00:31:14.400 –> 00:31:27.629
Dave Boyce: So when you… let’s just… I mean, I love it. Let’s start there. That’s current for you. I think it’s topical for everybody. Obviously, you’re way out in front of it from a research standpoint. So when you say agent, how do you define that?

276
00:31:28.030 –> 00:31:35.460
Tim Sanders: Very good question. Let’s get into that discussion. So, it’s a very nuanced question, right?

277
00:31:35.460 –> 00:31:39.230
Dave Boyce: you’re gonna have to have some definition, doesn’t have to be everyone’s definition, but you’re gonna.

278
00:31:39.230 –> 00:31:48.170
Tim Sanders: Yeah, let me define it, and let me talk about it as being on a gradient. So an agent, technically, is a complex system that has the capability to reason.

279
00:31:48.320 –> 00:31:49.350
Dave Boyce: Plan?

280
00:31:49.430 –> 00:31:52.970
Tim Sanders: Access tools, and ultimately, take actions in the real world.

281
00:31:54.030 –> 00:32:02.769
Tim Sanders: That’s what an agent… fully realized agent can do. However, agentic means to take an action. Now, the action can be taken in the background, like.

282
00:32:02.860 –> 00:32:13.469
Tim Sanders: chatbot co-pilot is what we call the low level, but it’s the beginning of the gradient. You say, wow, a chatbot is an agent? Yeah, you run a deep research session on OpenAI? It’s agentic.

283
00:32:13.470 –> 00:32:28.290
Tim Sanders: It’s making decisions for you, and the report is aging. Now, that doesn’t mean it’s put into production, you’re not putting that on your LinkedIn page, because you might have to check for hallucinations, but here’s how we think about it. We think that agents are on a gradient.

284
00:32:28.290 –> 00:32:46.110
Tim Sanders: And the gradient is moving up as time goes by, so at the bottom is the current chatbot co-pilot. Think of that as a prediction machine, like this great book back here behind me. It takes information you have, produces information that you don’t have. However, the human in the loop has to decide to put it into production, so its value creation’s limited.

285
00:32:46.110 –> 00:32:59.890
Tim Sanders: We’re seeing intelligent assistants coming along. ServiceNow has some, Copilot Studio, you can build some. These intelligent assistants have some of that scripting that we all know from robotic process automation. You give them a little bit more autonomy in your workflow.

286
00:32:59.890 –> 00:33:07.480
Tim Sanders: What we’ve seen recently, though, is the rise of task agents, whether you’re talking about Intercom’s VIN or even Agent Force, and they actually complete a task.

287
00:33:07.510 –> 00:33:11.139
Dave Boyce: Like, actually deliver something in the real world with some guardrails.

288
00:33:11.660 –> 00:33:15.539
Dave Boyce: And are we multi-step tasks yet, or not? This is just a single task?

289
00:33:15.540 –> 00:33:20.100
Tim Sanders: Process agents next. So, process agents, you think about a process like, go win a customer.

290
00:33:20.100 –> 00:33:20.420
Dave Boyce: Right.

291
00:33:20.420 –> 00:33:35.140
Tim Sanders: launch a marketing campaign, and to be honest, I see very few of those end-to-end process automations today, unless they’re augmented by robotic process automation bots. Like, I have seen organizations, and UiPath, I saw their research.

292
00:33:35.140 –> 00:33:54.410
Tim Sanders: If you have a strong foundation of bots that are 100% right, and they are, because they’re just RPA, and you put a little layer of agents on top for all the edge cases, you can do some end-to-end automation and deal with the error rate. It’s not bad, especially in low blast radius use cases like sales, where, like, there are, you know.

293
00:33:54.660 –> 00:33:56.410
Dave Boyce: One of the ways I’ve seen that

294
00:33:56.570 –> 00:34:01.500
Dave Boyce: described is… you said a system of bots.

295
00:34:01.870 –> 00:34:02.420
Tim Sanders: Yeah, yeah.

296
00:34:02.420 –> 00:34:04.700
Dave Boyce: But maybe as a system of…

297
00:34:05.410 –> 00:34:09.039
Dave Boyce: Agents with an orchestrator agent sitting on top of it, or…

298
00:34:09.040 –> 00:34:24.309
Tim Sanders: Well, what’s different is a bot isn’t based on a large language model. A bot is like good old-fashioned machine learning with scripts, and it works 100% of the time. There’s zero hallucination. However, it only applies to a limited number of situations, so the agent’s probabilistic.

299
00:34:24.310 –> 00:34:45.900
Tim Sanders: It can kind of move dynamically, but it also has compound errors. So if it has a 2% hallucination rate and takes 4 turns, it’s 60% because of the compounding factor. RPA’s zero, but it… so I think what we’re seeing now, Dave, is a combination of the two with, as you say, an orchestrator that’s kind of figuring out where to plug it in. And so that’s kind of coming. I’m seeing that the…

300
00:34:46.060 –> 00:34:50.179
Dave Boyce: Would you RPA with the orchestrator, or are those two separate things?

301
00:34:50.670 –> 00:34:53.129
Tim Sanders: Yeah, Orchestrator seems to be its own product.

302
00:34:53.449 –> 00:35:04.260
Tim Sanders: And you’re seeing a lot of it, like, there’s MuleSoft, there’s Maestro, there’s a variety of these. At G2, we’ve been covering it, it’s new. But it’s coming, we’re gonna see a lot more of that.

303
00:35:04.260 –> 00:35:18.949
Tim Sanders: But what we’re going to see in the next few years, though, is the system of agents. And the system of agents will dynamically create their own agents, they will do their own reinforcement learning in environments, they will deploy on their own, and we will react to them after the fact.

304
00:35:19.020 –> 00:35:29.359
Tim Sanders: And that’s coming, and that will create, you know, brute force opportunity for companies that can embrace it. And to kind of finish this analogy, Dave, the way I think of it is like Waze to Waymo.

305
00:35:29.690 –> 00:35:47.840
Tim Sanders: So when you… when you think about chatbots and intelligent assistants and even some task ages, it’s like having the Waze app on your phone. It’s very democratizing, you don’t need to know a city to get around, because it’s a suggestion machine, it gives you turn-by-turn, but your hands are on the wheel. Waymo is different! You get in the backseat, Jesus takes the wheel.

306
00:35:47.840 –> 00:36:03.050
Tim Sanders: And it is entirely transformative. It is a multi-trillion dollar industry as a result. When I talk about agents, I talk about that nuance, but to be very square, right now we’re around intelligent assistant task agent in this progression of dependable technology.

307
00:36:03.050 –> 00:36:04.650
Dave Boyce: That’s pretty low.

308
00:36:05.210 –> 00:36:26.669
Tim Sanders: We’re getting there. But we’re seeing incredible results even with the task agent. I mean, I was just reviewing some of those for the study. I mean, we’re seeing… we’re seeing sales taking 20% of the cost out of prospecting. We’re seeing… we’re seeing marketing taking 40% of the cost out of content production. It’s just amazing what even a lowly task agent can do.

309
00:36:26.670 –> 00:36:28.000
Tim Sanders: in 2025.

310
00:36:28.350 –> 00:36:32.300
Dave Boyce: So let’s explore that a little bit. I’m gonna start with the,

311
00:36:33.410 –> 00:36:44.419
Dave Boyce: you know, to get everybody’s attention, let’s start with where’s it breaking? Like, where did we overextend and think that we could hand things off to machines, and we learned, – that ain’t working?

312
00:36:47.710 –> 00:36:54.020
Tim Sanders: Well… I’ll give you some… if you’re asking me that question, I think the… I am.

313
00:36:54.020 –> 00:37:02.380
Dave Boyce: from experience, too, but yeah, like, let’s start with where it’s breaking, and then we can back up to, kind of, where’s it working, because you do have some success, it sounds like, as well.

314
00:37:02.840 –> 00:37:12.629
Tim Sanders: Yeah, so, let me just show a couple… no screenshots, please, folks, this is unpublished research, but hey, I want to show some love to Hard Skills Exchange and show you some stuff that hasn’t been put out yet.

315
00:37:12.630 –> 00:37:23.390
Tim Sanders: Love it. Again, we did a large study. This is the G2 August 2025 survey. We started with 4,600. We screened it down to… we screened it down to a couple of thousand, then we cleaned it to 1,500. It’s global.

316
00:37:23.490 –> 00:37:28.169
Tim Sanders: And we’re seeing the use cases largely be software development and customer support.

317
00:37:28.170 –> 00:37:31.190
Dave Boyce: Sure. So customer support right now is showing.

318
00:37:31.190 –> 00:37:39.560
Tim Sanders: about an 80% containment rate, which is great. When you compare that to call centers at 70% at 5 times the cost, it’s really working.

319
00:37:39.580 –> 00:37:58.159
Tim Sanders: the way to think about customer support is even if it breaks a little bit, in most situations, the customer support agent is providing the customer an answer they weren’t getting. They’re, like, stuck in self-service hell, or stuck in a chatbot saying, I can’t give you an answer, you know, call customer service. And so, it’s like… it’s like… it’s like 20 years ago when we went to VoIP, when we started to have those

320
00:37:58.220 –> 00:38:15.100
Tim Sanders: automated call center trees where, you know, it didn’t ring forever, it answered on the first call, but you had to navigate some menus. I mean, that wasn’t fun, but it was better than not being able to talk to somebody. So I find customer support to be one that, even when it makes a mistake, the pain is so great for both the enterprise and the customer.

321
00:38:15.100 –> 00:38:25.150
Tim Sanders: the damage is minimal. We call that blast radius in agents. Blast radius. The other way to think about this, Dave, fascinating insight I got from my friends at Harvard doing research.

322
00:38:25.820 –> 00:38:29.459
Tim Sanders: They study what’s called retraction latency. In other words.

323
00:38:29.650 –> 00:38:42.059
Tim Sanders: When a machine, like an agent, makes a mistake, how long does it take with new instructions for it never to make that mistake again? And the answer is, you measure it in milliseconds. But humans, the answer’s different. For sales, it’s 28 days.

324
00:38:42.060 –> 00:38:55.120
Tim Sanders: For the average employee, it’s 42 days, okay? So it’s like, we tend to think about the blast radius, but what we don’t think about is how quickly you can roll that back and fix it with agents. So that’s a fascinating point of view. So to answer your question straight up.

325
00:38:55.130 –> 00:38:58.179
Tim Sanders: The lowest breakage is in research and BI.

326
00:38:58.770 –> 00:39:15.540
Tim Sanders: So when you’re thinking about automating research, it’s never put into production. There’s always an analyst in the loop. There’s absolutely no risk to it. That’s why you can see here, it’s a strong use case across industries. In some cases, like, like healthcare, it’s the number two use case, right? Where it’s breaking is software development.

327
00:39:15.540 –> 00:39:20.970
Tim Sanders: Okay? I mean, everybody’s excited about vibe coding, but man, it only gets you from 0 to 1.

328
00:39:21.140 –> 00:39:21.820
Dave Boyce: Hmm.

329
00:39:22.260 –> 00:39:41.559
Tim Sanders: there’s just a lot of rework going on now with vibe coding on the back end, and you’ve probably read some stories, our friend at Saster, Jason shares the best, where he got his codebase deleted. Got his codebase deleted by an agent. So I see that as the one, where the breakage has been the biggest, even though the enthusiasm is the greatest.

330
00:39:42.480 –> 00:39:50.499
Dave Boyce: You had sales development at the very bottom. I don’t know what to read into that positioning, and I didn’t understand all the dimensions.

331
00:39:50.500 –> 00:39:56.909
Tim Sanders: You’re not… okay, so, so let me bring that back up, let me bring that back up a little bit for you here, and,

332
00:39:56.910 –> 00:40:12.610
Tim Sanders: Let’s do… I’m gonna go Sheldon Cooper here. It’s a little reference for my friends that watch Big Bang Theory. I’m gonna go Sheldon Cooper here. I like to segment survey results because, you know, you got all kinds of players here, so I kind of look at this a little bit. You know, in sales development.

333
00:40:12.640 –> 00:40:28.139
Tim Sanders: you know, when you kind of take a look at where it does better, you know, it’s not dead last in mid-market in SMB. It’s only dead last in large enterprise, but we… 40% of our end was large enterprise, because that’s what we wanted to focus on in the study. Let me just give it to you straight.

334
00:40:28.900 –> 00:40:40.909
Tim Sanders: So, so agents can reduce the number of players you have in a function. So, so customer service agents means you need less customer service reps. In sales, you need less salespeople.

335
00:40:42.300 –> 00:40:57.480
Tim Sanders: Sales leaders are empire builders. You tell a sales leader you’re gonna go from 30 to 60 employees, they won’t implement it. So there’s a real drag on adoption at the sales leader’s standpoint, not because the tech doesn’t work, but because the tech is, like, really disruptive to the endowment they have.

336
00:40:57.480 –> 00:41:03.949
Tim Sanders: for armies of humans. So it’s like, prove it to me, and I just got off a call with a sales organization, a big, large webinar.

337
00:41:03.950 –> 00:41:21.340
Tim Sanders: I can’t tell you what the… how big the resistance level is. You can show them, hey, I can reduce cost per lead to pipe by 20%. I can speed up lead gen to pipe, by 40%. Now, like, I don’t care, I don’t believe it. Not in my organization, not in my market, not for our customers.

338
00:41:21.340 –> 00:41:40.239
Tim Sanders: Why? Because the implication is, I’m going to eliminate SDRs as a function, and then focus more on a split between junior and senior AEs. No, they don’t want to do that, because it just runs counter, to their endowment around people aggregation, or hoarding, or whatever you want to call it. And I’ve been in sales a lot, so I’m not slagging on anybody when I say that.

339
00:41:40.240 –> 00:41:43.130
Dave Boyce: Totally. Well, let me push on that a little bit.

340
00:41:43.540 –> 00:41:55.889
Dave Boyce: let… just for the sake of argument, let’s eliminate the empire builder kind of motivation from the argument. Let’s just assume that that is true for almost all sales leaders, but we’re just going to eliminate it from the argument and.

341
00:41:55.890 –> 00:41:56.830
Tim Sanders: Sure, sure, sure.

342
00:41:56.830 –> 00:42:00.640
Dave Boyce: We’re just gonna be very rational about, you know, evaluating this.

343
00:42:02.120 –> 00:42:07.329
Dave Boyce: So, that’s kind of number one. Number two, you’ve already established that in a support use case.

344
00:42:07.760 –> 00:42:25.150
Dave Boyce: robots can solve the problem X percent of the time, like, without a, you know, without a human, and they can do it… and we already know, theoretically, they can do it in all languages, they can do it 24-7, they can learn… and you just established that they can… they can, correct their mistakes quickly, they can learn quickly, and I’m kind of close.

345
00:42:25.150 –> 00:42:35.019
Tim Sanders: 80%. The containment. It’s the best metric, is the containment rate, and yeah, 80% versus 70% call center. And by the way, 65% offshore call center. Just a little side of that.

346
00:42:35.020 –> 00:42:51.130
Dave Boyce: So good. So, in theory, I should be… and that’s a human-to-robot interaction, right? And it’s succeeded. In theory, I should be able to translate that over into the… to a human-to-robot interaction around sales development, like qualification, early-stage sales cycle stuff.

347
00:42:52.340 –> 00:42:52.719
Tim Sanders: Why not?

348
00:42:52.720 –> 00:42:53.330
Dave Boyce: fruit.

349
00:42:53.570 –> 00:43:00.880
Dave Boyce: I’ve heard the argument, not from a, empire-building standpoint, but I’ve heard the argument

350
00:43:02.160 –> 00:43:17.200
Dave Boyce: that eliminating a human there is, isn’t gonna work. It’s gonna be disconcerting. Like, I have a motivated buyer coming in, and she needs to be able to talk to a human, not a robot. How would you respond to that?

351
00:43:18.550 –> 00:43:36.219
Tim Sanders: Well, that’s a segmentation. I mean, I would use… I would use demographics. I would say, how old is this person? What demographic are they in? Are they in one of the SEC states? I’m just kidding. That’s actually an Oklahoma, Texas joke for you. You have to segment that answer to have a real view into what we’re talking about. So the question is.

352
00:43:36.220 –> 00:44:01.209
Tim Sanders: When I talk about sales automation, I’m not talking about the actual Zoom call. The Zoom call is always done by a human. The interaction’s by the human. There’s a ton of work to get a person on a Zoom call. That’s where the opportunity is, okay? So all of the persona enrichment, all of the cadence enrichment, all of the back and forth scheduling enrichment, no, we don’t want to talk to a human to set a Zoom meeting. We’re a real buyer. We want… but on the Zoom meeting, we do. So you’ve asked your

353
00:44:01.210 –> 00:44:02.359
Tim Sanders: Really good question.

354
00:44:02.380 –> 00:44:14.449
Tim Sanders: If we set aside the empire builders, and I was asked this question, you, you are right on here. Okay, guy says I’m not an empire builder, I still don’t do it, tell me why. Here’s the reason why. Let me tell you a little short story.

355
00:44:15.610 –> 00:44:32.539
Tim Sanders: Let’s say you’re a parent of a kid in college, or let’s say maybe they’re even, you know, freshmen, and you are micromanaging them, and you don’t trust them to take a trip to Cancun for spring break, and they’re like, you don’t trust me, you don’t trust me.

356
00:44:32.820 –> 00:44:42.189
Tim Sanders: But you have another one that’s 40, and you did that show with them, and they don’t come home anymore for Christmas, because you kind of broke up with them in college. They stopped calling you, and…

357
00:44:42.570 –> 00:44:44.310
Tim Sanders: The pain is so great.

358
00:44:44.610 –> 00:45:01.290
Tim Sanders: you will give the new one, the second child, more trust than you ever imagined, because you saw what happened when you didn’t do the other. The reason I bring that up is a concept that I’ve been working on for my next book, and it’s called Trust Gaps, and I studied it for cloud computing. Cloud computing trust gaps were big.

359
00:45:01.400 –> 00:45:14.740
Tim Sanders: For 15 years, researchers studied how much money companies lost by not moving to the cloud because they didn’t trust it for whatever reason. Answer’s $44 billion a year over 15 years, okay? If you think the trust gap’s bad.

360
00:45:14.880 –> 00:45:19.779
Tim Sanders: look at what it’s gonna be for agents. It’s gonna be huge. So here’s the answer. The answer is…

361
00:45:20.010 –> 00:45:22.600
Tim Sanders: The pain is not great enough in sales.

362
00:45:22.700 –> 00:45:46.870
Tim Sanders: For us to trust it. The pain is great enough in customer service. When you look at customer service debt, anything, even mistakes, is better than nothing. When you look at software debt, especially around updates and making shipping deadlines, the pain is so great, and it’s breaking, there’s… it’s still the fastest growing category on dollar for dollar. That’s why API spend is 80% software coding.

363
00:45:46.870 –> 00:46:05.099
Tim Sanders: So, so follow the pain, because where there is pain, there will be begrudging trust. So sales just doesn’t feel enough of that pain right now. They don’t have the recruiting bottleneck. They don’t have the cost per containment, pain, that customer success is having. I mean, my God.

364
00:46:05.100 –> 00:46:18.409
Tim Sanders: Call centers, $10 on the average. $10. An agent, on the average, is somewhere in the neighborhood of $1.80 to $3. So, follow the pain, and you will find adoption. Sales doesn’t have it yet.

365
00:46:18.470 –> 00:46:29.210
Tim Sanders: Marketing, a lot of those CMOs don’t do it because they come from a brand background, and for them, the blast radius is just… it’s unacceptable. It’s a one-way door. And I always tell sales leaders, like.

366
00:46:29.210 –> 00:46:43.950
Tim Sanders: you say you’re worried about sending bad emails that aren’t well-constructed. I’m like, have you read the stuff coming out of fill-in-the-blank machine learning platform that’s doing that automation already? So, anyway, it’s the pain issue, really, if you get past Empire Builders.

367
00:46:43.950 –> 00:46:49.670
Dave Boyce: For a half a second, maybe this is gonna persist, I don’t know. It depends on how much pressure we get from…

368
00:46:49.780 –> 00:46:51.349
Dave Boyce: You know.

369
00:46:51.980 –> 00:47:02.040
Dave Boyce: how much pressure enters boardrooms from watching, you know, companies like Lovable and Replit and Cursor go, you know, bonkers with no salespeople? I mean, we should have taken the cue from,

370
00:47:02.190 –> 00:47:14.739
Dave Boyce: from Atlassian and Dropbox and DocuSign, but we didn’t, you know, the no salespeople kind of… all of the fastest-growing software companies since 2003, all of them, like, every single one.

371
00:47:15.170 –> 00:47:19.539
Dave Boyce: has been PLG. Every single one. Slack, DocuSign, Atlassian.

372
00:47:19.540 –> 00:47:22.270
Tim Sanders: All the way up to Enterprise, and once you get to…

373
00:47:22.460 –> 00:47:37.829
Tim Sanders: If your investors think you’re gonna… well, I gotta tell you, Windsorf salespeople took me to a Cowboys game, and they’re doubling the sales force every quarter. So, no, Windsorf got to a certain level, and now they’re gonna get to… and I already know a level.

374
00:47:37.830 –> 00:47:39.569
Dave Boyce: Oh, you’re saying it’s fastest growing.

375
00:47:39.570 –> 00:47:40.289
Tim Sanders: There’s a ceiling.

376
00:47:40.290 –> 00:47:41.060
Dave Boyce: enterprise, and.

377
00:47:41.060 –> 00:47:47.269
Tim Sanders: TLG has this… there’s a ceiling. I mean, it’s… it’s one thing to be zero salespeople to 200 ARR. Well, that’s not true.

378
00:47:47.330 –> 00:48:06.400
Tim Sanders: 120… if I’m gonna get really wonky and look at a piece of unpublished research I just read that studied Replit, lovable, even mid-journey, etc, your ceiling’s around 120 if your TAM is around 150x that. So it’s like you take the TAM and you do some math, and there’s a ceiling without salespeople that PLG can get you through.

379
00:48:06.400 –> 00:48:19.890
Tim Sanders: go ask Aaron Levy about this at Box, and he’s going to tell you, you want to be a billion-dollar run rate company? Guess what, folks? You’re going to build sales. You’re just not going to build it from mid-market and SMB. So that’s the answer, there. It’s nuanced.

380
00:48:19.890 –> 00:48:22.960
Dave Boyce: But just coming back to the pain question,

381
00:48:24.050 –> 00:48:28.250
Dave Boyce: Does, you know, the market is back to growth. Like, the market is demanding growth.

382
00:48:28.380 –> 00:48:39.390
Dave Boyce: That’s right. No. Like, we’re done with that. That was 2022. We got… now we have at least 20 of our points coming from growth, and… That’s right.

383
00:48:39.400 –> 00:48:59.069
Dave Boyce: And are we gonna throw money at it again the way we did in 2019, 2018, 2017? Question mark. If we don’t, then there could be some pain. I don’t know, this is a hypothesis, Tim. Like, the pain is not high enough in sales, because sales has a recruiting pipeline, and they have a budget, and they have a budget, but what if we pulled the budget from sales?

384
00:48:59.530 –> 00:49:03.749
Dave Boyce: Like, what if… what if people said, yeah, I want you to grow what you actually can’t hire?

385
00:49:04.180 –> 00:49:06.860
Tim Sanders: Amazing point. So,

386
00:49:07.340 –> 00:49:23.140
Tim Sanders: we were actually able to analyze at the organizational level, and this is G2 review data, so I can’t show it to you, because, you know, we want to respect privacy. But… but we were able to, behind the curtain, I’m able to kind of correlate companies, and then what I do is I use, hiring data to correlate,

387
00:49:23.140 –> 00:49:39.099
Tim Sanders: the change in open job recs, or announcements of layoffs, or rumors of hiring freezes across scrapable places, like using Grok on X, there is an absolute correlation between a company suddenly changing the sales hiring practices and them adopting agents. So, to your point.

388
00:49:39.100 –> 00:49:40.249
Dave Boyce: There you go.

389
00:49:40.250 –> 00:49:43.929
Tim Sanders: To your point, once they’re told you can’t backfill.

390
00:49:44.410 –> 00:49:47.340
Tim Sanders: Once they’re told, we want you to top grade.

391
00:49:48.220 –> 00:50:08.380
Tim Sanders: you know, they’re gonna… the pain will become great, so right now, you’ve hit the nail on the head. Go build an army, go hire all the SDRs you’ve ever… you ever wanted. That’s going on even at PE-backed companies, but if that changes, and we started to see it at large enterprise, where you’d see, like, some of the big companies, Meta’s a great example, or even Salesforce, what’s going on right now.

392
00:50:08.380 –> 00:50:14.730
Tim Sanders: Where you can kind of manipulate your stock price by your cost… your capital efficiency by letting people go.

393
00:50:14.760 –> 00:50:32.600
Tim Sanders: once that starts to move downstream, you’re gonna see more urgency at the sales level to backfill with something. So if I tell you, Dave, you can’t backfill with 5 SCRs, then you’re gonna take a really hard look at agents to be that backfill, because it’s coming out of OpEx, it’s not coming out of

394
00:50:32.600 –> 00:50:53.880
Tim Sanders: fixed cost headcount. So I do think that’s a hack. And again, we played this game before in tech with cloud and premise, so when they cut all the capex, everybody just started putting it in the cloud, because you could use OpEx, and that was a big driver. 2008 recession was the number one driver of cloud adoption, not managed cloud, private cloud, and improvements in privacy.

395
00:50:54.210 –> 00:50:54.800
Dave Boyce: Well, let’s assume…

396
00:50:54.800 –> 00:50:56.500
Tim Sanders: Very good point that you’ve made.

397
00:50:56.500 –> 00:51:08.699
Dave Boyce: I mean, let’s just… for the audience, let’s just lean into that version of the future, because I actually… I think some version of that is going to happen, where backfill budgets are going to get pulled, the pain is going to rise, and…

398
00:51:08.770 –> 00:51:19.210
Dave Boyce: revenue leaders are gonna have to figure out how to do more with less. You know, so they’re used to building an army, now they gotta build a clone army. What you’ve said…

399
00:51:19.340 –> 00:51:25.890
Dave Boyce: Well, what do we… let me just back up. I don’t want to lead the witness.

400
00:51:26.090 –> 00:51:36.689
Dave Boyce: I’m that person now. My budget just got pulled, I still have to deliver the same number. Actually, I gotta deliver a growing number, and now I gotta learn how to work with robots. What do I need to know? Like, what…

401
00:51:36.840 –> 00:51:42.289
Dave Boyce: Tell me how to step into that new reality as a revenue leader. I’m not a technologist, I’m a revenue leader.

402
00:51:43.520 –> 00:52:00.670
Tim Sanders: Start with the business problem and work backwards. Focus on your ability to very clearly express goals, and organize enough context of what good looks like, what success looks like, best examples for every use case you might want to automate, because right now.

403
00:52:00.670 –> 00:52:08.700
Tim Sanders: Context is king, whether it is in using a large language model or in programming an agent. So let me break this down one step at a time. So.

404
00:52:09.070 –> 00:52:12.200
Tim Sanders: Where agents are going, and they’re going there very quickly.

405
00:52:12.400 –> 00:52:17.070
Tim Sanders: Is to, fulfilling well-expressed, defined goals.

406
00:52:17.240 –> 00:52:17.940
Dave Boyce: Yep.

407
00:52:17.940 –> 00:52:32.650
Tim Sanders: So if you can… and it’s almost like managing… if you’re a CRO, it’s almost like managing your regional vice presidents. That’s what it’s always been about. Help… help describe the goal, but give me the right KPIs so that I can kind of break that down and see what success looks like.

408
00:52:32.650 –> 00:52:45.420
Tim Sanders: Organizing context is really important. Let me give you… well, I’ll give you a sales example. So I was looking at a sales organization that’s outperforming their rivals in an industry by 40% on agentic efficiency.

409
00:52:45.420 –> 00:53:08.330
Tim Sanders: The one thing they seem to do that no one else did is they did all the upfront work to gather all the gone call data from their best sales calls that were the most correlated to closing a deal, and put that into natural language transcripts to train an agent on. They looked at all the gone calls in particular that got a Zoom meeting with a person further up in an organization. They also gathered

410
00:53:08.330 –> 00:53:13.740
Tim Sanders: all of the emails that had high conversions say in trying to set a Zoom call for that use case.

411
00:53:13.740 –> 00:53:18.029
Tim Sanders: So when you take the time, and there’s actually contractors you can bring in to do that.

412
00:53:18.280 –> 00:53:21.879
Tim Sanders: you gotta really get them under an NDA, of course. But what they’re doing

413
00:53:21.930 –> 00:53:38.860
Tim Sanders: is they’re carefully studying the best converting emails, the best converting gong calls, and creating a rich context document that the agent trains on. You don’t need to be a technologist to do that. You need to know two… you need to be Rick Rubin. You need to know the goal of the record, and what good looks like.

414
00:53:38.860 –> 00:53:51.819
Tim Sanders: And Rick Rubin is my example, and if I use one thing just to illustrate this, because I think it’s in… I think it’s, like, very empowering to people on the call today, is that when you think about AI and inclusive of agents.

415
00:53:51.820 –> 00:53:57.649
Tim Sanders: what it’s doing for the first time is it’s decoupling prediction from judgment. So in the past.

416
00:53:57.650 –> 00:54:13.300
Tim Sanders: the expert did both at the same time and made a decision, and then they had rules of thumb that ran their organization, and the problem is rules of them, they go out of date, and you find out, you know, when it’s too late. So what’s happening now is the machine can make the prediction, and then the human being passes judgment.

417
00:54:13.550 –> 00:54:20.979
Tim Sanders: And that produces very dynamic data decisions that you can code. You can change the code, as I mentioned earlier, immediately. So the point is.

418
00:54:21.130 –> 00:54:26.300
Tim Sanders: You don’t have to become a tech expert as a sales leader. You need to get your together.

419
00:54:26.380 –> 00:54:38.389
Tim Sanders: With respect to organizing your greatest hits. And a lot of times at organizations, we really haven’t done a good job of that. Some have done a very good job of it, especially if they have a strong partnership with ABM, because that’s ABM’s

420
00:54:38.390 –> 00:54:46.570
Tim Sanders: thing is to go find out the best emails that convert and make sure that becomes the only emails that are sent, right? So I think that’s the answer.

421
00:54:46.570 –> 00:54:52.570
Tim Sanders: There’s one little thing, too, for everybody watching. Go Google, because I don’t have time to look up the link, just Google

422
00:54:52.790 –> 00:54:54.570
Tim Sanders: 30% rule.

423
00:54:55.060 –> 00:54:56.040
Tim Sanders: Harvard.

424
00:54:56.400 –> 00:55:02.359
Tim Sanders: You’ll get a great video, and it basically says you only need to learn 30% of a technology to master.

425
00:55:02.360 –> 00:55:05.649
Dave Boyce: Boom, I didn’t lead the witness, but you got there anyway.

426
00:55:05.650 –> 00:55:09.940
Tim Sanders: That’s right, you just need to learn a little, make you dangerous. Fascinating.

427
00:55:10.520 –> 00:55:18.219
Dave Boyce: So, you said something at the very top that I think anyone listening here who’s interested in this topic should really key on.

428
00:55:19.040 –> 00:55:21.549
Dave Boyce: You said, start with the business goals.

429
00:55:21.840 –> 00:55:41.409
Dave Boyce: And, like, I gotta have… this is so important. Then you went to what correlates with the business goals, which is, like, let me analyze, kind of, the activities that will actually get me to the business goals. But even knowing what the business goals are, and having them written down, is necessary in order to deploy a robot army. It’s necessary in order to deploy a human army, too.

430
00:55:41.410 –> 00:55:42.040
Tim Sanders: That’s right.

431
00:55:42.040 –> 00:55:45.690
Dave Boyce: Sometimes assume that it’s just kind of, like, already exists in the ether.

432
00:55:45.690 –> 00:55:46.180
Tim Sanders: Yes.

433
00:55:46.180 –> 00:55:52.190
Dave Boyce: It’s gonna get transferred through, kind of, tribal knowledge from generation to generation, and we didn’t.

434
00:55:52.190 –> 00:55:52.630
Tim Sanders: Yes.

435
00:55:52.630 –> 00:56:11.249
Dave Boyce: time to document a theory of the case. But if we have a documented theory of the case, then a robot can actually follow, can actually optimize for that. And then the next level of, well, what does correlate to good results? And so… but that first thing, like, have a theory of the case. If you’re going to manage a robot army, you need to know how to give them instructions.

436
00:56:12.040 –> 00:56:30.039
Tim Sanders: Because here’s why this is important. The concept of Parkinson’s Law, right? I don’t know if you heard Parkinson’s Law. Parkinson’s Law, I actually think I still have it. Oh, yes, I do! Another graphic! This one you can screenshot. This is a game changer. Let me tell you why generative AI has not produced results, but…

437
00:56:30.080 –> 00:56:32.870
Tim Sanders: agents using generative AI do.

438
00:56:33.060 –> 00:56:52.859
Tim Sanders: Parkinson’s Law says that humans will always expand to fill the allotted time, meaning even if I’m more productive, it doesn’t mean I produce more, it means I have more time that I can spend watching TikTok videos, going to do things I want to do, walking my dog, or going to extra Zoom call meetings. Deadlines drive human results, not

439
00:56:52.860 –> 00:56:53.650
Tim Sanders: tools.

440
00:56:53.760 –> 00:57:13.109
Tim Sanders: And Parkinson’s laws, like, been proven by so many studies. So that’s why, even though generative AI can make you faster at coding, it doesn’t mean you ship more code. Deadlines make you ship more code, but if you change deadlines because you gave people productivity tools, they will riot. Agents don’t do that. They grind.

441
00:57:13.280 –> 00:57:28.080
Tim Sanders: They are unaware of deadlines. They don’t take PTO, they don’t get caught up in Slack conversations, they decline 100% of Zoom call meetings, so you can actually forecast an increase in velocity 99% correlated to the efficiency of the agent.

442
00:57:28.350 –> 00:57:35.259
Tim Sanders: Okay, now that changes everything, and that’s why agents are such an exciting opportunity. So when you tell an agent.

443
00:57:35.260 –> 00:57:50.979
Tim Sanders: I’m going to give you a goal to produce X number of Zoom meetings, and you’re going to point at that goal, but I’m not gonna give you a deadline. I’m going to say get there as quick as you can. My research says they’re going to surprise you 85% of the time and show up early.

444
00:57:51.420 –> 00:57:58.100
Tim Sanders: Assuming you pick the right platforms. And by the way, there’s a question that I need to address here, because I keep referring to this research.

445
00:57:58.770 –> 00:58:06.530
Tim Sanders: Yeah, question is, how are you collecting usage and manufacturing self-reporting? Yeah, all surveys are self-reporting. I wish we could actually take a look at the books.

446
00:58:06.530 –> 00:58:19.770
Tim Sanders: from private to public and see where they actually spent it. So, we did a lot of work at G2 to take a large cohort of 4,600 and kind of screen and clean it down, to a smaller cohort of around 1,500 that feel like

447
00:58:19.770 –> 00:58:25.909
Tim Sanders: They’re in the know and have low noise, but it’s self-reporting, just like G2 review data is.

448
00:58:26.480 –> 00:58:41.589
Dave Boyce: So, let me… I know we gotta hand it back, Julia. Let me, let me just summarize a couple of things that we heard for all revenue leaders here. Number one, get out of your own head on empire building. That is not the game. The game is delivering results. So get out of your own head and solve…

449
00:58:41.680 –> 00:59:01.059
Dave Boyce: and solve for delivering results. Number two, and if you have to kind of fictitiously raise the pain to get yourself into that headspace, great. Number two, agents aren’t only running Zoom calls. Like, that’s not the only way to think about agent as a customer-facing role. You can also think about agents as taking on tasks that are behind the scenes, that set up the Zoom role.

450
00:59:01.060 –> 00:59:07.529
Tim Sanders: Everything up to the Zoom call, and maybe after the Zoom call, you can automate, but the Zoom call needs to be a human. That’s my… I believe you.

451
00:59:07.530 –> 00:59:13.990
Dave Boyce: And then number 3, make… do the intellectual work to develop a theory of the case. Like, un… like…

452
00:59:14.090 –> 00:59:17.430
Dave Boyce: Make sure that you can write down what good…

453
00:59:17.600 –> 00:59:29.409
Dave Boyce: what the objectives are, and then maybe even do some research on what good looks like, so that you can actually task your agents to accomplish the results that you’re targeting. Thank you so much, Tim. Amazing. I want to spend more time with you, brother.

454
00:59:29.410 –> 00:59:32.439
Tim Sanders: Well, then let’s do it, man. Let’s do it.

455
00:59:32.750 –> 00:59:33.709
Julia Nimchinski: In that last session.

456
00:59:33.710 –> 00:59:35.800
Tim Sanders: Julia, I mean, back to you. Yeah.

457
00:59:35.960 –> 00:59:44.890
Julia Nimchinski: Thank you, Tim and Dave. The community loves it. Lots of comments, questions, but yeah, the main question, how can we support you? Where should our people go?

458
00:59:45.830 –> 00:59:52.789
Tim Sanders: I put it here, if you could put it in the chat, just come see me on LinkedIn. I’ve got a newsletter, I publish everything all the time, that’s all I care about. Dave?

459
00:59:53.240 –> 00:59:56.829
Dave Boyce: LinkedIn also, and I just released a book. I know Tim has written.

460
00:59:56.830 –> 00:59:57.660
Tim Sanders: Brilliant.

461
00:59:57.810 –> 00:59:58.769
Dave Boyce: But my book…

462
00:59:58.770 –> 00:59:59.210
Tim Sanders: you.

463
00:59:59.210 –> 01:00:14.450
Dave Boyce: No, it’s like 2 weeks out, it’s called Freemium. It’s all about the rise of automated GTM, and now moving to Agentic GTM. Also, and the flip side of that is self-service. We have a long ways to run on both of those trends, and I’d love for you to read Freemium.

464
01:00:15.690 –> 01:00:16.570
Julia Nimchinski: Amazing.