Cut Twice, Measure Once: Why It’s Better to be Fast than Perfect with Greg Ceccarelli
Episode Summary
Product teams spend tons of energy planning — and almost none of the planning survives contact with the market.
Greg Ceccarelli (CPO, Spec Story / former CPO, Pluralsight) has a phrase for it: "cut twice, measure once." Get things out into the world. Most decisions are reversible. The market will teach you more in a quarter than a year of planning ever will.
Greg and I got into: — Why AI flipped the "longest pole in the tent" from engineering to decision-making — How Spec Story is trying to close the gap between "we decided this a Zoom ago" and actually building it — Why $5/hour might be a better pricing model than anything seat-based — And why truly authentic content can’t ever be automated.
Guest
Greg Ceccarelli - Chief Product Officer at Spec Story, an AI-first startup building tools to make AI coding easier and safer. Before Spec Story, Greg held product leadership roles at Pluralsight (CPO), GitHub, Dropbox, and Google (you can find this without a link), and spent years as a consultant at Alixpartners and IBM.
Show Notes
What does it mean to cut twice and measure once? Greg Ceccarelli, CPO of Spec Story, has built his career moving fast through IBM, Google, GitHub, Dropbox, and Pluralsight — and he's learned that the biggest bottleneck in building software isn't writing code anymore. It's making decisions. Greg and host Tom Noser dig into how AI is rewriting the economics of software development, why SaaS seat pricing is broken, and what it actually takes to build shared context across a team. Plus: the real challenge of distribution in a polluted content landscape, and why Greg believes asking questions is his single greatest professional skill.
In this episode, Greg and Tom discuss:
Moving fast vs. planning — Greg's "cut twice, measure once" philosophy, why most decisions are reversible, and what happened when he pushed back on a private equity firm's annual planning process
AI and software development — How AI agents are compressing implementation time, changing the economics of software, and flipping the traditional "longest pole in the tent" from engineering to decision-making
Spec Story and Stoa — How Spec Story started by preserving AI chat history for developers, and why Stoa is now focused on capturing collaborative meeting context so teams can move from decision to implementation faster
SaaS pricing — Why seat-based pricing is past its expiration date, and how Stoa's $5/hour model is designed to remove friction, align with value delivered, and eliminate the token-opacity problem
The future of SaaS — Headless software, API-first systems, and whether agents will make traditional UI obsolete
Distribution and marketing — Why distribution has gotten harder, not easier, why authentic human content outperforms engineered content, and what questions every founder needs to keep asking about their customer
Core competency — Greg's answer: asking questions, and the compounding value of learning velocity over specialization
Transcript
Tom: [00:00:00] Why is it better to be fast than to be accurate? What does AI mean for SaaS and software development and how can AI strengthen collaboration? These are some of the questions I asked Greg Ceccarelli, chief product officer at Spec Story, an AI first startup that builds tools to make AI coding easier and safer. Before joining Spec Story, Greg worked for companies like IBM, Google, GitHub, Dropbox, and Pluralist, where he was chief Product officer. Greg worked as a consultant, investment banker, marketer, product guy, and now he's our guest on the Fortunes Path podcast.
Tom: [00:00:47] Greg, it is great to see you. Thank you so much for being on.
Greg: [00:00:51] Thank you for having me, Tom. It's a pleasure.
Tom: [00:00:54] I know we met recently and I heard you describe yourself as a cut twice measure once kind of person. I've never heard anybody use that description before. Tell me what you mean by that.
Greg: [00:01:10] I've sort of always been this way. So, you know, literally, if it comes to like hanging a picture. Right. I kind of tend to want to get it up there and maybe want to get it up there fast. And what that means is, you know, I'll actually start to eyeball where I'm going to put the nail in and put the nail in and then adjust the picture instead of measuring it. I don't know, super comprehensively, I guess. And so I guess what I was using that phrase in context of is, you know, when it comes to building a startup and getting validation on your ideas and understanding, if you know, whatever you're doing has value. I like to not just throw things at the wall, but I'd much prefer to move fast than slow and be very measured, especially in this day and age. So when I said cut twice, measure once, I sort of meant, you know, we can think strategically all day about what to do and of course we should do that. But it's better to get things out into the world and have them sort of meet contact with reality. And so that's sort of just a governing force.
Tom: [00:02:25] Well, I love your analogy about hanging a picture. It's exactly the same way that I hang a picture. Which means I end up with holes in the walls if I ever want to take the picture down. I've got holes in the walls. But I'm interested. Have you worked in big companies or have you primarily been in smaller start ups sort of organizations?
Greg: [00:02:44] I have worked in the largest of large. So I started my career at IBM and IBM Global Business Services, and then I worked in a consulting capacity for seven years after that. Primarily a company called Alixpartners which is a turnaround and restructuring company first and foremost, but they have all sorts of different service lines. And then I moved on from that to work at Google. So it's a pretty big company.
Tom: Yeah. Awful.
Greg: Maybe one of the biggest, I think whenever I joined it was about 80,000 employees. It's probably closer to 200,000 at this point. And then a tour of duty down, down the stack. So Google was the largest company by far that I've ever worked at. But I've worked at Dropbox, GitHub seed stage startup, a midsize business, Pluralsight, and now a very, very teeny tiny startup Spec Story.
Tom: [00:03:39] How well does cut twice measure once work in a place like Google or IBM?
Greg: [00:03:46] You know, that's a good question. I mean, I think it completely depends on what part of the business you're in. And so like, if you're like in an incubated new bet, for example, it's going to work a lot better than if you're, you know, working in the performance zone, sort of cash cow part of their business, right? So for Google, that would be something like search or ads, right? I didn't work in those functions. And I don't know if necessarily my the way that I worked was governed by that, by that statement in those places. I mean, you have to kind of adapt to the environment you're in. But I think that there is definitely a lot of truth in that. Like, you know, maybe it's cliche at this point, but like how Jeff Bezos talks about, you know, there's only N numbers of one way door decisions, right? And a lot of companies spend so much time fretting over decisions that are easily reversible. And so it's like a mantra. I think it's really important to realize that, you know, most decisions are reversible. It's better to get sort of feedback on a product feature that you ship, something that you're trying to do faster than not. And yeah, again, I'm, I'm not saying that that mantra is a panacea for everything, but it just sort of governs my my, my desire to, to sort of move more quickly than not, if possible.
Tom: [00:05:15] I had an old boss who I got crossways with where he was a planner and his anticipation was, is that you were going to be doing a lot of planning. And I said to him, well, I'd rather do my work than plan my work. And he and he blew up and he goes, it's like, it's the same thing. And I think for a lot of people they don't understand that difference between planning and doing particularly in big company environments. Sort of have you ever encountered a similar misunderstanding with somebody?
Greg: [00:05:51] I mean, of course I can tell a small anecdote. That was interesting and sort of funny at the same time. So whenever I was at Pluralsight, Pluralsight I was the chief product officer there. And we were under the ownership of Vista Equity Partners. And when it comes to planning there's a very rigorous sort of Vista best practice, if you're familiar, like Vista was famed for buying primarily SaaS businesses and then sort of instilling a set of operational best practices. And one of the most famous ones is their product development practice. And what that entails is a very robust sequence of steps where for every single thing you might want to put on your roadmap, you know, you have to validate it with, you know, ROI analysis, like, you know, feasibility, all sorts of stuff. I won't go into like the extraneous detail, but just imagine like almost like the biggest waterfall type product development process you'd ever seen. And I think for some businesses that like Vista owns or owned historically, that might have worked really well because the pace of change was dramatically different than, you know, the environment that I was in whenever I was there and we were going through the planning process, I think for 2023 ish, I think. And I just sort of threw up my hands and said like, look, like I'll do a six month roadmap, but there's no way that we can actually plan out a full year. It doesn't make sense for, you know, the change that we're going through.
Greg: [00:07:33] And I think that might have been one of the first times that they had ever had someone push back like that especially at an executive capacity and not just sort of go through the motions, but it turned out that, you know, for all of our product development teams and they're like, it's like 60 squads, something like that. It was like a breath of fresh air because the amount of time that went into the actual annual planning that, you know, was really an input into like funding effectively. It was like to validate, hey, you know, we're going to, you know, give you more money even though it's like one pocket to the other. You know, how can we be confident that there's going to be a positive ROI on this? And of course, you can never be fully confident from a product development roadmap. But my, my whole point was like, you know, we're going to learn so much more in these, in this next quarter that anything that we spend time planning for now is almost like wasted effort. And I understand having a little bit of a view out, especially for, you know, a company that sells enterprise software, you know, the, the product development cycles are traditionally longer. You know, and sort of match to get your sales cycle. It's like almost, like takes a quarter to do anything really. But yeah, that's, that's one of the, a good example, I think.
Tom: [00:08:54] So you're working in an AI related startup right now and I just have to ask the cliche question. You've said like in these days and pace of change, etc.. What's your point of view on how AI is changing software development?
Greg: [00:09:14] That's such an expansive question. Yeah, we can talk for a while about that. Okay. Well, I mean, I think it's probably clear by now to anyone that's paying attention that you know, the amount of leverage that you can get using AI agents that can actually implement code to specification is enormous and is only increasing in capability. It wasn't like that a year and a half ago where, you know, the frontier models were much less advanced and the harnesses that they ran in, you know, didn't have the same tooling and polish that they do now. But, you know, relatively speaking, historically, you know, the longest pole in the tent had always been sort of implementation time, right? Like, so I am in product and I come up with something that I want to build and I validate that there's a reason to do it. And I have a good hypothesis for it. I prioritize it. And then of course, I'm, you know, delegating it to teams of engineers to actually implement. And of course, all the interfaces between those different teams have to get resolved. And, you know, it has to actually be built to spec, which is usually pretty fuzzy because there's handoffs and there's just a lot of time, right? That's why, you know, the entire DevOps software engineering, intelligence sort of market sprung up to try and say, hey, like we should measure the throughput and quality of, you know, what we're building because it's, or has historically been the largest cost center for any software technology company that gets sort of flipped on its head when all of a sudden, you know, assuming you have a fairly good idea of describing what you want.
Greg: [00:10:53] And that is a very large assumption. You can actually, with speed and quality, implement extremely quick relative to the past. And it changes the economics of building software dramatically. Even if you're spending a lot of money on tokens, it's still a drop in the bucket relative to the size and cost of traditional engineering teams. And so it changes everything, right? It changes everything. And it changes the economics of software construction and the way that things get priced and, you know, brought to market and customer expectations. And you know, what were existing moats, right? It used to be you're going to pay for some SaaS software because, you know, the transaction costs of you building it yourself are much higher than buying it for $29 a seat. You know, even if it's sat there sort of underutilized, which most SaaS does, most software does. Right. So I'll start there. But why don't you interject so I don't just go on a monologue.
Tom: [00:11:58] Well, there's I think it changes everything as well. So, you know, there's the what's his name? The guy who runs he's, he's suing or being sued by Elon Musk right now. He runs OpenAI. Sam Altman yeah. Where his thing about the solo unicorn, you know, solopreneur who creates $1 billion business. Actually, that's, that's not a bad point of view. I mean, it's very kind of psychotic asocial, you know? But I do think that what he's the type of business that he is envisioning there does seem possible. My, my sense is these machines are sort of passably bad and but they get to that level of passably bad really, really fast. And they're they're like acceptably bad is better than mine. You know, my first draft of acceptably bad is like unreadable. But what they are able to do looks like something meaningful and without analysis, it passes that initial test of just like, oh, this looks reasonable. And to me, that's one of the reasons why they're so dangerous is overused with AI, but potentially deceptive or misleading. Because they pass that initial smell test. It's only when you start to sort of really, it's hard work to dig into one of these things and discover the flaws. And so I don't write software code. I, you know, I've, I've coded, but I can't tell you like, oh, what is the way it wrote this? That just is not going to scale that once this gets to like, you know, 100 users, there's some process in here that is so inefficient in the way it was written that it's going to crash everything.
Tom: [00:13:56] And I think that's a, we made those mistakes when we built all of our software by hand. And it's not like those are new mistakes. So are they going to get repeated in the AI, or is the AI just sort of like going to be smart enough that you avoid a lot of these problems of scaling that SaaS businesses have had in the past? But I think that your point of getting smart through iteration, which was what scrum was supposed to be, but never really was. And then being able to get those in front of people to have actual customers react. And it's so much cheaper now, so much easier to have something that a customer can honestly react to than it used to be. I remember paper prototyping. And it's like, oh, you would fake it on paper and then show it to somebody and it worked. Okay. And then Figma and all the other stuff, but none of it's nearly as convenient as what these things do now to where you can actually shape an idea. So I'm, let's talk a little bit more practically about how you're using the tool. So you're in several businesses. One of them is called Stowa. And Stowa had a chance to recently try out. Tell me a little bit about Stowa and what you guys are trying to achieve.
Greg: [00:15:21] Yeah. So our company is called Spec Story and we have effectively like two product lines the original spec story extensions. And then we have a spec story cloud is focused on the individual. And what I mean by that is very early on there was a pain point that was very acute about people adopting tools like GitHub copilot and then cursor, and then the terminal agents like cloud Code, Codex, cursor, CLI, droid, you name it, where they wanted to preserve their chat history. The sequence of interactions that they had with the agent, like all the turns of conversation. And so we, we, we built products to make it super easy and efficient for people to just install an extension and get clean markdown in their repository so that they had a record of their chats. It was sort of amazing to me that these vendors didn't make it easier to extract. Over time, there's been sort of a little bit more standardization, although there's not a unified protocol for you know, how a conversation, which is a sequence of messages between a user and an agent gets stored. And so we did the sort of unglamorous work of mapping to a more standard file format for all of these different vendors, so that people didn't have to do that, and in some cases, reverse engineering. You know, like the, the storage location of the chats. And I think there's like simultaneously been a very large sort of conceptual push into the value of context because anyone who is using AI to produce code, not building AI powered products, but actually using LLMs to write code, realizes that your effectiveness is largely constrained by, you know, the detail and first, the prompt that you can provide in terms of how you're going to govern that conversation within the context window.
Greg: [00:17:26] But then further buoyed by the the sort of right context being provided in addition to your prompt. So what I mean by that and like very tactically is, okay, there was like phase one is prompt engineering. Like, how do I write a precise enough prompt so that the LLM knows, you know, what the next best tokens are to predict to satisfy the aim of my query. The second bit is, well, okay, how do I extend that and also give it access to perhaps tools that it can use so that it can actually search the internet? Or like there's a few files that I want to specifically point it to. So it has that particular context. So it doesn't just randomly search or grep for, you know, things that are irrelevant, which would not only just burn tokens, but probably lower the likelihood of a good response. This idea of context is that's a very tactical version of it. I think more interestingly, like, you know, if you think about the way that enterprise companies operate there are natural like context routers in those companies, like executives, for example, generally have broader context about, you know, what the business is trying to do, what the customer pain points are, you know, what the constraints are, yada, yada. And so that was a long way of getting into what STO is, which is, look, if spec story saved your AI chat conversations that you're having between you and your agent, you know, there's this other really important context where product decisions and other sorts of things happen, which are meetings, right? Like if you're not working in a development capacity and sending an agent to crank out lines of code, chances are you're discussing what to do next or reviewing what you've done with another person.
Greg: [00:19:16] And, you know, one of the main sort of places you do that is perhaps on a Zoom call or a team is meeting or in Slack or, you know, these other asynchronous places. But what we wanted to do, what we are doing is provide the sort of same conversation capture for teams that are trying to collaborate and work with agents to produce code and you know, a meeting like environment. And so that's, that's the bet right now that we're pursuing because we think that you know, there's already a pretty, what's the right word? There's already an expectation that you're going to have meetings at work. Yeah. Right. Yeah. And most people that have meetings at work spend a lot of time in that meeting discussing things, planning, not doing. But there are tools and capabilities that you can bring into the meeting and your context to actually get work done during meetings, and then have the side benefit of having, you know, all of your transcripts captured the decisions you've made, anything that you've built in that meeting being available and syncable to wherever you need it. And so in a nutshell The sort of pain that we're trying to close is the gap between, hey, I decided this a Zoom ago and I kind of forgot.
Tom: [00:20:40] Right.
Greg: [00:20:40] Right. What it was. And now we're off scratching our heads saying, like, what are we going to do? And it's just a lot of wasted time.
Tom: [00:20:47] That's interesting. There's a lot to comment on that, but you made me think about some advice I got from my dad, who was an entrepreneur many years ago about when you're in the meeting don't say anything until the very end. And then at the very end, summarize the meeting for everybody. And when you summarize it, you get to create the outcome, the agenda from that meeting that you want. So the summary doesn't actually have to be an accurate summarization of what everybody said and agreed to. It's your take on what you think the, the, the next right thing to do is, and I do think that that's actually a super powerful because people don't remember, you know, you have the meeting and you don't really you're in the context of the meeting and then the meeting is over. You don't remember what the meeting was about. But if I can sum it up in a bow at the end where it sounds at least related to what we talked about, I can have my way. And I do think that that's something that I don't necessarily want to give over to an AI. So I also think that there's a we talk about like wasted time in meetings. I don't think communication is literal.
Tom: [00:21:59] So these, these machines don't interpret emotion. They don't think, they don't have emotion. It's just, you know, I don't have to go into it, but it's, you know, ones and zeros and but we're not that way. Like we're, we're more, we have thoughts and feelings. They're kind of the same thing. And we our thoughts are the way that we express our feelings or we describe our feelings and our feelings, our thoughts we can't express. And these machines don't have any of that. But I think those things are still important contexts in person to person communication. So to put that in a more practical terms, the session I did with Stoya was a lot of fun. And, but what it felt to me like was parallel play. So it's like when you see little kids developing in a sandbox, when they first get to know each other, they're each doing something in the sandbox, but they're doing different things. They're in the same sandbox, but they're all doing different things. It takes a while for them to develop where they're playing the same game. You know where I'm the Germans and you're the Americans or whatever. And so I feel like that and that may just be that I haven't played in store very much.
Tom: [00:23:16] And so that's, it's more a reflection on my ignorance than it is on anything about the machine. But it that to me, that's like when I collaborate with an AI that's kind of parallel play where I give it something, it tries to please me or it's like talking to my dog. Dog doesn't really understand, but it's trying to please me. It does something. And then if I have a positive reaction, it smiles and I'm kind of like, is that real work? Anyway, this is what this is something I'm sort of struggling with is like, does AI produce real work or is it fake work? You know, because so much of what I do in my work is actually fake work. Like you talked about that whole process with your private equity owners and they wanted you to jump through all these hoops before you could put something on your backlog. That to me is just fake work. You know, there is no way to absolutely know what the return on investment is going to be. Why pretend? Why not just take a limited risk, measure the results and get smarter? Anyway, so that's my struggle right now. Or one of them is, is this stuff even real work?
Greg: [00:24:33] I want to respond, but I want to actually ask you a question, which is like, what, what do you consider real work?
Tom: [00:24:39] Yeah, that's a great question. And the answer is that I don't know when it's happening. I only know after the fact. And so I can't tell the difference between the things that I'm doing that aren't going to have a positive impact on the outcome and things that won't have a positive impact on the outcome. So I'll make an example of like writing a poem. So if I write a poem, I have to sit down in a chair and actually write the poem. That's absolutely real work. A lot of the stuff that I write when I'm writing a poem won't end up in the poem, but I can't get to the end of the poem directly. I don't know what things not to write that won't end up in the poem. I have to do both before I get there. So I can say like, does that feel like real work? The crap that doesn't go into the poem. Yes and no. And then the same thing about like. All right, well, what about what inspires the poem? I go for a walk with the dog. You know, I have dinner on the porch with my wife. Yeah, those things are are background information that I'm not even conscious of. But some of them are critical to be able to write the poem. So I'm not really answering your question other other than saying.
Greg: [00:25:57] No, but you did. You did actually, because I'll summarize it for you. It's not the end of this meeting yet, but it seems like it seems like you're all of this activity, right? If you wanted to classify it a priori, it's sort of hard to put it in real work, not real work buckets, but all of the activity, like in the, like in the posterior, like looking in the rear view mirror is sort of helping you run towards like a sharper truth. Yeah. Right. Because like that final poem is the sharpest truth up until you revise it the next time you sit down to edit it, right. That you have. And I think one of the things about the Stowe environment that you experienced that, you know, we don't have, I don't purport to say that we have the right answer for is that it does have a lot of bells and whistles, which can be distracting. Like there's a lot of things that you can do. Some things aren't progressively disclosed as well as they could be. But the whole point of the environment, which brings together, you know, live video, audio transcription, sort of a shared workspace where you can simultaneously edit documents and also collaborate with an agent is, is built in order to help teams get towards More strongly wedded mental alignment. So while you can use the environment to, for example, rapidly prototype an idea that might feel like parallel play because two people are spinning two agents at the same time, and, you know, not everyone's actually like interacting at the exact same time together. The real goal is to say, well, hey, you know, we need to bring together the things that we would otherwise be doing on our local machines, right? Using Cursor, using Copilot, using Claude Code.
Greg: [00:27:55] And we also need the documents or assets, whether they be code or, you know, a spec document. And, and be able to not have to fiddle around the friction of screen sharing and trying to teleport into one another's machines and making it present and available so that we can build shared mental alignment as a team, trying to move, you know, to the next right outcome. And so like in our experience, that wasn't really exactly like how we used it. We were, you know, trying to kind of prototype. And I, I did, but it was, you know, without a lot of shared history, right? Because you came on and you met with Jake and I and, you know, you had to bring us up to speed rapidly in terms of what you wanted to try to do. Right. But for a trusted team that's been working together for a while, you know, and say there's, you know, on something that they're building like a big architectural decision, right? And someone's done some pre-work to start to draft out, you know, like a design for how they're going to modify the system. There's not a lot of tools out there that allow you to then poke and prod and sort of see where everyone else is in the context of absorbing this and then make modifications all in one place. And again, the goal isn't like, hey, we just want to add collaborative document editing and, you know, video context and agents all in one thing and call it a day.
Greg: [00:29:22] It's the real goal is, you know, if implementation time is the shorter pole in the tent, you know, how do we clarify the right set of ideas together as a team who are going to be responsible for delivering this thing faster? And ultimately, if you had to measure that, which isn't baked into our product today, but it's something that we think about all the time, it's like, what's the intent lead time to get to the sort of first working version of the thing now? Right? And what I mean by that is from the time we make a decision about, yep, we're committed to doing this to actually, you know, the first commit where you could actually see some evidence that it was produced. How do we shorten that gap? Because that gap has historically in product management been extremely long in a lot of cases, especially as the organization grows and expands? Like it takes a long time to go from decision to implementation, but I think that that's the new that's the that's the new like sort of war zone, for lack of a better term. Like that's where companies are going to get made. Like they're going to make it or break it. It's like we have all these great ideas. How do we, you know, prioritize them, activate them first, start implementing them. And then, of course, you know, release them and get feedback from the market and validation on if they're valuable or not. Right. Because like everything else is compressed. So that's what we're building towards as a vision.
Tom: [00:30:47] Well, let me react a little bit to the vision. So I'm involved in a project right now that's supposed to be collaborative and it's a large organization. And we spend a ton of time repeating ourselves. There isn't a shared problem space and each person who's participating in the project, their own thinking evolves when they're not in the meeting. And then they come back in the meeting and they're talking about their latest thinking and you don't understand the thinking they had last week, or I don't understand the thinking they had last week. And so if Stoa can fix this idea of, okay, we're all going back to our offices, we're all using our own agents, we're all putting crap into those agents that we assume is the context for the project. None of that's shared. It's all stuck on everybody's individual machine, and no one is like a librarian of context who can see everything in there and begin to make decisions about, oh, we changed our mind about this. Take that one out of the context. We put this one in, put that guy in. That would be enormously helpful to be able to particularly in large organizations if you could. Oh, when you say roadmap, this is what you mean when I say roadmap, this is what I mean. And there's this, the, the Stoa context knows that that this word has different connotations depending upon who says it, and can then educate the rest of the team about that. That to me, seems enormously beneficial.
Greg: [00:32:21] Yeah it does. It's a hard problem, obviously, but but yeah, I mean, like, you know, there's been a lot of like enterprise knowledge management solutions that have not necessarily purported to solve that problem, but have, I guess at least latently tried.
Tom: [00:32:38] Yeah. And they all feel like.
Greg: [00:32:40] And they all fail. I think one of the, the smaller, more discrete problems that we want to try and first crack is, you know, how can you make the context between humans that's very important and hopefully high alpha. Be made available to agents even in the first place. Right. And how can you make that like sort of durable? And what I mean by that is like one of the key differences between other meeting products that you might have used. And it is it's not a product designed to, you know, book a scheduled meeting for. It's more like room based where you have a space that can have N numbers of projects associated with it. These are some things you didn't see because you came in as a guest. And any session that you have, whether it is with an agent in store or if it's with a team member or multiple team members, all of the topics of discussion, you know, they get written to mark down in that project folder associated with that room. And so anyone can effectively search any of that context at any given point. Now, it doesn't solve the vocabulary problem, but it's like the foundation to being able to actually build and curate sort of more meaningful, like business company context over a time horizon, right? Because if you think about how Zoom works, right? You know, you jump into ephemeral Zoom. Yes. You might have an AI notetaker record meeting notes, but where does it go? It just gets sent to your email and then like forgotten about, right? In this case, that same, that same room or space that you had your meeting in two weeks ago has all the history, not just of the voice conversations that you had, but the interactions with any of your agents and all your, your project files.
Greg: [00:34:37] So, you know, the core vision is not fully realized, of course, like this is a new product, but the sort of building blocks and the primitives are in place such that you might be able to imagine a fairly, you know, near-future where assume you onboarding your onboarding someone new to your company, and you can basically summarize, you know, a sequence of decisions around a particular topic area that had happened very easily because you'd been using this product and you had all your sort of context loaded into it. That would be something extraordinarily unique that, you know, like from a value perspective, like, of course it's a big lift to say, hey, you know, stop using all these other things that you do and use this, but like, there will be companies that are, you know, born tomorrow that don't need the heavy weight office 365. Right. You know, Google workspace type environments because they're primarily interacting with markdown effectively with their agents to build stuff. And so it's more, it's certainly more niche in terms of its focus, but the idea is that your context should be made explicit, and it needs to be readable by agents because they work best when they have access to file systems. Now all that might change, but that's where we are now.
Tom: [00:35:55] Where we are right now. Yeah. It's interesting. So I know you guys introduced a kind of different pricing model for Stoa and it's this…
Greg: [00:36:07] Yeah.
Tom: [00:36:08] Yeah. Tell me about how that evolved and what was the reasoning behind that?
Greg: [00:36:13] Yeah, it's a good question. And it went through a lot of different iterations. And I guess we're still the jury's out on whether or not it will succeed. But the pricing for it is actually very simple. So you're charged by like effectively the number of meeting hours you use it. And so it's $5 an hour to use the product and on its face, you might think that like that could discourage usage, but if you do some of the math and we have a calculator that makes this like pretty obvious you know, the nice thing about that compared to a seat based model is immediately, if you're smart and you're like seeing, you know, if you're thinking about it just for like a minute, Like there's no there's no seats. There's no concept of seats. Right. So I can, I can add as many people as I want and into my organization. And then no matter if there's ten people on the meeting or 20 people on the meeting, although I think meetings with more than 20 people are probably more like webinars than actual meetings. Because I don't know what you would get done on them. It costs the same. And so economically, I mean, we wanted to sort of reduce the barrier to entry to try to the point where it felt like, you know, do I not even value $5 of my time? Right. Especially whenever I'm getting all this other stuff for free. To, to reduce the, you know, again, like the sort of trial syndrome. But then also align it with, you know, the value that the product's delivering meaning, you know, let's say you were a very small team and you didn't actually need, you know, to meet that frequently.
Greg: [00:38:02] If you start comparing the the sort of equivalent subscriptions that you need to sort of bundle together to get the same functionality for limited utilization. I mean, this just like wins night and day, but of course you don't position or sell anything purely on price. Sure. Right. That's not great. But but it but it had gone through a lot of revs like we did consider a potential sort of, you know, seat based plus like credit system. The reason why is because, you know, you have access to cloud code in a sandbox, right? And so it could be pretty financially sort of disingenuous for us to think that if we didn't charge credits for tokens, like people wouldn't abuse the thing. But I think a lot of vendors that are trying to add AI onto their sort of seat based pricing model or, you know coming into a lot of friction from a user adoption perspective, because there's all these different like credit models out there. We wanted to make it as simple as possible to understand. It's like, it doesn't matter if you use cloud code in the sandbox. However much you would have previously spent for it, it's like, just how long is that sandbox running? And that's like part of the $5, meaning you don't have to think about tokens or anything that are very opaque. And quite frankly, like when you're writing a prompt if you are paying per unit today, you don't even know how much that query is going to cost you until you know the results come back, right? Because it's going to go think for, you know, five minutes and consume, you know, 10,000 tokens. And then you and.
Tom: [00:39:43] You.
Greg: [00:39:43] Get the usage. So from a. Yeah. So from a customer standpoint, you know, it's, it's, it's like very simple, you know, it should be mapped exactly to like how much you use it. You don't like lose your, your dollars that you put in if you don't use it for a week. Right. And so that's, that's ultimately what we landed on again. The jury's out, but I think it's almost close to like paying for an outcome as you're going to get. Because we don't actually know what your outcome is, but, you know, it's, it's sort of like usage based pricing, but it's, it's like if you're actually having effective meetings and making decisions, like the outcome are decisions from meetings. And that's what we help effectively synthesize and classify for you.
Tom: [00:40:25] There's a lot I really like about your pricing model. One of them is that it it aligns very well with the, I'm gonna say kind of central analogy of the application, which is that this is a special room. It's a room where you can do things that you can't do in other places. And so we don't, as you said, have one person in there doesn't make much. Have two people in there, have 20 people in there, $5 an hour. The other thing is it seems so outrageously cheap. So to your, to your point about like getting people just to try it. That feels like absolutely at a, at a, just a try level. And then it also lends itself to different pricing tiers in the future related to stuff like the consumption of tokens. So, you know, up to 10,000 tokens in an hour. And like you said, nobody knows. You're like, you're going to ask one query and burn all 10,000. But that's not, that's just the way the AI's work. But anyway, I really liked it because I thought it would lead to adoption. And then ultimately I want people to get addicted to this room. It's just like, well, we save so much time and we get so much done when we're in the room. Let's all get in the room. And so we're using it eight hours a day continuously. And then it's just a part of doing business and I'm not even thinking about it.
Greg: [00:41:49] Yeah. That's the hope. That's the dream. Yeah. That's what we're shooting for.
Tom: [00:41:55] Anyway, I was, I was really impressed because I feel like SaaS pricing seems to me like it's all screwed up. I mean, we're great when we, when we first introduced this idea of seat pricing, it seemed brilliant. But I think it's, it's past its prime. It's gone past its expiration date.
Greg: [00:42:10] It's totally. And I think that's like a more existential question. I think a lot of, you know, existing, especially public market SaaS businesses are asking themselves because there had been like obviously an explosion in SaaS buying over the last ten years. And when you go and audit things like this as part of a larger organization, which I had the, I don't know, I don't know what the right word is displeasure of having to be a part of my role. Pluralsight. What you realize is there's like kind of a power law of, you know, the SaaS subscriptions that you have and what gets, you know, almost 100% utilization and what gets like close to nothing. And of course, this despite everyone's best effort to have good procurement processes and reviews and whatnot, there are just so many unused seats that exist out there. I mean, like order of magnitude. When we went through an audit past company, it was like in the tens of millions of dollars a year for a 2000 person company. And I'd say that like outside of like 6 or 7 vendors and we had about 290. Yeah. You know, nothing was actually used, relatively speaking. And like the, the name vendors you might expect are kind of like most of the systems of record, right? Like the sales forces and gongs and Githubs and Slack's of the world, the things that the business actually ran on. And so I think a lot of those companies are obviously sort of safer than not.
Greg: [00:43:45] But for those that, you know, were like a convenient sale or, you know, whatever, there was a small need that that sort of didn't really materialize after the initial purchase, and it was just easier to continue to renew because no one was like, you know, pulling out the magnifying glass. A lot of companies are going to have to, I think, get much more competitive with their pricing because there is, of course, a refrain. I don't know how it's playing out in large organizations. It's like, well, you know, Claude could just build me this in a day. And I think that's a dangerous approach, but not for everything. Because historically, even for those main systems of record, you were generally trapped into their sort of vanilla business process model, right? You had to customize this thing to do all these different things. And, you know, if you can build the thing that you need with, you know, just the 2% of features you actually use, as long as it's not some megalithic, you know, CRM or something like, I mean, a lot of, a lot of people might do that, but it's not most companies core competency to build their internal, you know, system. So I think there's a lot of talk, but we'll see how that actually plays out in reality about, you know, is that does that actually happen?
Tom: [00:44:58] I mean, that's, that's a fun topic to is like future of SaaS. I saw something recently I'll ask you this and then I want to ask some marketing questions, but I saw something recently about decapitation. Have you heard about this idea of taking the UI off of everything and yeah, yeah.
Greg: [00:45:18] Like headless, headless, headless systems that are just API based.
Tom: [00:45:23] Yes. And it, what's your thought on as a product person and as someone would you call yourself a marketer or would you say you have a marketing disposition?
Greg: [00:45:37] I need to have a bigger marketing disposition. Tom. Distribution turns out to be, you know, one of the hardest things now when there's so much noise and a compressed time to build everything. But I like to think a lot about positioning. I have a particular fascination with that. But on to your question about like, you know, decapitation and headless SaaS. I think that is the preamble to the not entirely figured out use case, but obviously burgeoning. One of, you know, what's, what's the UI need to be? Yeah. Right. And so I think the most recent example was like Salesforce saying, hey, you know, like we're going to go effectively headless with agent force and, you know, like that coming, I don't know all the details about it, but I obviously saw the soundbite and the news headline. I think that the point is it's like, you know, for the past like 20 years, we've been building UI flows for humans because humans had to interact with screens and click buttons and input data into forms and then query it and do analytics on it and get it out. And if an agent has the right context, meaning it is somehow hooked in through MCP or APIs to your existing systems of record, especially if they're fragmented systems of record.
Tom: [00:47:02] Right?
Greg: [00:47:03] Which most are. Despite vendors best efforts to try and either buy or build all the integrations they possibly can to reduce that problem, then like, you know, do we need all these screens or can we just ask the, the agent, you know, whatever that interface is, whether it's text or voice to synthesize sort of the result we might have needed to, you know, munge all this data together, put it in a warehouse and query the answer for it. Right. And so that's a very simplistic view of things. But like, you know, there needs to be a persistent store of information for businesses and there needs to be a way of accessing and modifying it. And if you don't need screens and you can just interact or interrupt through an agent, then it seems to me reasonable that a lot of these existing SaaS that have very complicated, sometimes less so UI would be thinking about a different UX, right? The experience that most people want is probably much closer to talking to ChatGPT, like wrongly or rightly, right? Like I can just talk to ChatGPT. If it had context, you know, it could give me the answer. It could synthesize an HTML, some chart or graph, like, you know, all these canvases that exist in the products already do it. And yeah, I mean, I understand the trend. I don't know how, I don't know how it ends ends. Right, right, right. But I understand the trend.
Tom: [00:48:25] Yeah. It is, it's it's a fascinating trend in that there's, you know, like human computer interface was always dependent upon what are the limitations of the computer. And they still have limitations, but they act as if they don't. That to me is like, AI is this magic trick Of I understand you perfectly and I can respond to your desires, which is like a wonderful personal assistant. And so it does seem reasonable that that would become the dominant way that we interact with machines is by talking to something that seems like I'm talking to a person. It's Hal 9000. Hopefully, hopefully not psychotic. But you had said something about needing to understand distribution more and getting more attuned to marketing. Tell me a little bit more about that.
Greg: [00:49:23] Oh, I mean, so I think that like marketing is a lot of things, but the thing that's nearest and dearest to my heart is trying to make the value of whatever you're producing for the benefit of someone else legible.
Greg: [00:49:42] Right. Like how do you make it legible? Meaning understandable in context at the time of need. And so there's of course, all sorts of different types of ways that you market for different reasons, right? You know, brand marketing is for the constant awareness, right? Whereas, you know, demand gen might be for conversion event and, you know, the traditional funnel. But at the end of the day, you know, depending on what the intent is, if it's, hey, I'm browsing the internet, I'm just trying to like learn about something and, oh, you know, wow, I should probably know about this thing because this person wants to build my awareness that might lead to a, you know, a future purchase event or conversion event, like whatever the thing is being marketed like that, that value needs to be legible and that sort of moment. If someone is, you know, higher intent because they've already done their research and are deciding, hey, I'm going to buy this productivity collaboration software, right? Like normally you need to have a very sharp, legible sort of value proposition in whatever form that might be, whether it's copy on your website or an ad that they see or, you know, some organic type of thing that they're reading. Right. Because you've been doing content marketing and figuring that out because it's never static is like a really interesting problem that I haven't until recently spent a lot of time on.
Greg: [00:51:08] But in my past, I certainly had and I think it's actually the hardest problem now, and not to be super cliche because like, everyone's like, yeah, distribution is always the hardest problem, but like, it's gotten way harder because there's so much noise in the market, right? There's so many and most channels are completely polluted just with an explosion of content, right? So, you know, if you're trying to build awareness or whatever on LinkedIn or x slash Twitter, you know, the feed is just a spigot that never stops spitting out, in some cases, a lot of slop, in some cases more useful information. And then, of course, there's algorithms that control what everyone sees. And you know, it's just harder to stand out. Yeah. And it's harder to continue to be relevant because every single day there's just a, you know, a new shiny thing to try out, right? And the pace and acceleration has just picked up dramatically. And then, you know, more traditional channels have just also become more expensive because they're more saturated. So it's always about figuring out the unique way to get into the place where you expect your ideal customer to live and make your value legible. And it's like, you know, a whole big game. I'm just interested in it, but I'm not, I'm not necessarily good at it, but I'm very interested in it.
Tom: [00:52:33] I have a bookshelf that I'm looking at right now that's got about, I don't know, 25 boxes that has each box is 40 books in it. So I've got like 2000 books in my basement. A book that I wrote and produced and stuff. And so distributions. Horrible. I was in Los Angeles visiting my daughter and back then, I guess I had more courage, but I approached a group of attractive young women and I showed them my book and I said, well, you know, what would you think about? Where would you look for this book? In this bookstore? And one of the women says, oh, we should ask, you know, Phyllis, she works in publishing. And Phyllis takes the book and she says, oh, this is adorable. Try to get it in bookstores. And that was her, you know wisdom in the publishing business is like, oh, try to get your book in bookstores, which is a lot harder than you might think, or at least it has been for me. I actually think that's absolutely the wrong answer. I think the right answer for me is go to live events where people can meet me and sell the book one at a time to individuals. And so that's like go to a pug meetup. The book is written by a pug. And so go to a pug meetup and meet people who enjoy these dogs. And they'll see the cute dog on the cover of the book and maybe want to talk to you. And I feel like, okay, so how does that translate to spec story in store? Well.
Greg: [00:54:05] That's a good question.
Tom: [00:54:06] Talk to everybody who is a who, you know, in your personal network who's a developer or works with developers and you know, it's just, it's one damn person at a time. I yeah, yeah, I just to your point about like the pollution of social media and everything else like that, it's like, well I used to work for a social media company and they had somebody who was their They're like content whisperer. But they hated his message. His message was there is no way to automate content. You just can't do it. Content is a creative process. It's something that has to be legitimate, not feel legitimate, but feel authentic and feel legitimate for people to react with it. And you can't turn that into a formula. I'm sure no one has stopped trying. But I think he had a really good point. Which is that there was, there's something about the human example, like a reason I'm not terrified of like I can tell the difference between things I write and things Claude writes when it's imitating me. I don't know if readers can. And so I haven't experimented with that yet. And there are a lot of implications for what you're for trying to get distribution, for trying to get attention. If it's possible to automate a voice that's an authentic voice is now automated. And that changes it. Or is it really always going to be piecework? Is it always going to be by hand?
Greg: [00:55:48] Yeah. I mean, I kind of tend to agree with that person you named. I, it's interesting like this, this is like less about like, you know, company distribution and more about sort of experimentation with just personal audience growing on like LinkedIn, for example, over time. And I was like, looking back at some of the analytics of things that I posted. And, you know, some of the ones that had by far like the most engagement or repost or shares or whatnot were sort of like off the cuff thoughts that I had that I, you know, actually wrote. Well that I think it's, it's kind of like, you know, I wasn't necessarily trying to engineer an outcome or like thinking super hard about how it would resonate, but probably in whatever the algorithmic distribution, magic, magic that happens felt like, you know, a human wrote this and it was like legitimate and, you know, got served and, you know, reacted to others and they weren't like provocations. They were like, you know, just like bundles of experience about particular things, some version control, which is like a weird topic, but others were just about like how I'd used, you know, agents over the last year before all, everyone, any, any time ever talks about is just how they use agents because that's now like all that's even on. And so it's a little bit of like, you know, newness, newness and relevance at the time.
Greg: [00:57:23] But yeah, I kind of agree like with that person and what you said, I think the landscape is always changing. And so it's very hard to automate something because there's going to be some amount of deterministic rules in that that then gains relevance in a changing environment. Like, I mean, you kind of have to be reacting to it. So I don't know. I mean, at the end of the day, I think when it comes to marketing, there's like a lot of questions you're trying to sort of ask and answer and all the answers to those things are always changing too. It's like, who is this thing for? What do they believe? What pain do they already feel? You know, what category do they think your thing belongs to? Always changing. You know, what do they have to understand before they actually truly want it? Like, why would they use your thing instead of just doing nothing? Right? Like the status quo argument. And then like, why now? Like, why are they going to use it now? And you have to be able to like dynamically answer those questions with your value proposition. And again, it's like, how do you, how do you automate all that? Because people's thoughts and beliefs and needs and wants and desires and status Madison. Constraints are always changing too.
Tom: [00:58:34] Yeah. So I think.
Greg: [00:58:36] Yeah.
Tom: [00:58:36] I liked your your thumbnail sketch of the questions you have to answer to be able to bring something to market there. And the thing is, you could ask Claude that and it would come up with reasonable answers, but none of them would be informed. They would because it doesn't talk to anybody. It doesn't, you know, it hasn't conducted research other than what's already available on the web. And so this, this is the good enough versus. Right. And but so that first cut of it that Claude does, maybe your work style is to say, I'm going to go take that first cut and then go test it. And then I'll come back to the AI and I'll give it what the results of my testing until it's a to modify based upon that, that new data. Or you could say, I'm not even going to ask Claude. I'm going to do the first draft and then do research on that first draft, and then come back to the machine at some point and ask it to tweak whatever it is that I did. But anyway, it's like I love, I love the challenge of like trying to understand people trying to understand why they're doing what they're doing. And so I'll end with this question, which I meant to ask earlier. I think everybody has a core competency like I feel about myself as I'm an educator, I've got a master's in education and I still think about teaching and learning as like central to what I do. Some people are salespeople, some people are accountants. What would you say is your core competency?
Greg: [01:00:24] Asking questions.
Tom: [01:00:27] I like that one too.
Greg: [01:00:30] I'll give you a slightly longer answer. Because I think it's born in, I don't know, I think this is somewhat intrinsic. I suppose you could learn how to be more of it, but I've always just been like a very, like, curious person myself, like trying to figure out how things sort of work.
Greg: [01:00:48] And the longer answer is like, you know, through not just introspection, but also inspection, like asking questions, especially like if you're using LMS or agents today can really help clarify your thinking, which then leads to like a higher learning velocity. And I guess another way of saying that is like figuring out how to be adaptable, I think is a core competency to me. And that might sound like fluff, but I think I've always been someone who's been extremely eager to take on, you know, different types of opportunities and challenges and in different environments, purely because I thought that it would help me improve my learning velocity, which, you know, learning and knowledge sort of compound more than anything. And I think we're sort of seeing the fruits of what 40 years of internet texts can turn into when it comes to distilling it into a model that can, you know, do inference at the, you know, whatever the speed of the tokens are these days. But I never wanted to have like a single specialty. Right. Because I don't know, I think there's like some Heinlein quote about like specialization is for insects. And there was like a point in time where like specialization was very vaunted. But I've always preferred the more generalist track because I think the benefit of being able to sort of connect the dots through you know, different experiences, but also different domains is extremely valuable. I think that's where innovation comes from. Right. Innovation comes from being able to say, oh, this works over here in music theory. Why don't we apply it to business? Like, I'm just making that up. You get the point, right?
Greg: [01:02:36] And, and so, yeah, I mean, I think figuring out how to adapt in different environments as a skill set is something that maybe even from my earliest days of being a consultant, getting plopped into situations I had no business being in really helped me value not just variance of experience, but sort of the quality and ability to sort of connect the dots later on. And so all that's made possible by like being able to ask the right sorts of questions, not necessarily of others, but also of yourself to figure out what you sort of have to, to do next.
Tom: [01:03:17] Most important question there is, is what do I do next?
Greg: [01:03:21] Yeah. What will I do next? That's a great.
Tom: [01:03:24] Question. It is. It is.
Tom: [01:03:26] Greg, it was really fun talking to you. I appreciate your time very much. This is great.
Greg: [01:03:32] Yeah, it was a pleasure and a lot of fun as well.
Tom: [01:03:44] The Fortunes Path podcast is a production of Fortune's Path. We help service and technology businesses address the root causes that prevent rapid growth. Find your genius with Fortune's Path. Special thanks to Greg Ceccarelli for being our guest music and editing of the Fortunes Path podcast. About my son, Ted Noser. Look for Fortune's Path book from Advantage Books on Fortune's path.com. I'm Tom noser. Thanks for listening and I hope we meet along Fortune's Path.