The Defaults Your Power Users Set

0:00 Markus Bernhardt

a technical team brings an HR tool in for recruitment. as the agentic piece improves, you've basically signed a new recruitment policy.

the moment they update in the background, they have updated your recruitment policy. So for those who think, this is a very technical discussion, I don't care what buttons in the back end of Codex or Claude Code are pressed, No, no, no. this is very much frontline,

Customer, client facing and internal work.

0:38 Sam Rogers

Welcome to Signals and Subtractions. I'm your host, Sam Rogers. Who in the organization is actually setting the AI defaults that everyone inherits? Not who owns AI, not who signed the contract, but who decided what the thing does when nobody's watching for everyone who comes after them. I guess many of you can name that person, but I'd guess that nobody voted for them.

My co-host today is Dr. Markus Bernhardt. And it was actually in talking with him that made me decide to turn the newsletter I'd been writing every week for a year or so into this very show. We've written a handful of articles together and like to get together to inspire each other's work from time to time. So it's great to finally have him join as co-host. Markus is a strategist for the future-ready AI-powered workforce.

A Fortune one hundred consultant, advisor, board member, international keynote speaker, and editor of the popular Endeavor Report. coincidentally, he runs Endeavor Intelligence. Markus, welcome, so glad you're here.

1:42 Markus Bernhardt

Really good to be here, Sam, and thank you for the kind and warm intro.

1:45 Sam Rogers

Of course. it's really perfect timing to have you join the show. So most people advising organizations on AI, they're advising about adoption, right? What to buy, what to pilot, what to stand up. But you spend your time on the part that comes after the buying of things, which always, you know, masquerades is progress, but as I like to say, like buying is never progress for anyone except the salesperson. so

Determining what is actually implemented and how to how to find our way to what the organization is becoming is what I love jamming with you about. So today is not so much about what to adopt. It's about who set the terms that everyone else lives inside.

You'll get one signal one subtraction from each of us Markus, you're up first. I'm so glad to have your signal. What do you got?

2:36 Markus Bernhardt

Fantastic, yes. And this is the interesting bit, and that's why I like my job. I like looking at what use cases and what endeavors would be useful in an organization, in a team, in a function. And how do we make them a reality with the current tools? And how do we know where the boundaries are, where reality moves to hype? that's that's the section of realism that I like to work in. And so my signal is that very often.

It is the more technical people who bring a tool into the organization. And when those technical people bring the tool into the organization for their use, they set the terms that that tool will operate under. And very often, then

When the tool rolls out further throughout the organization, those permissions just remain what they were right at the start. No one reviews them ever again, and they just got set at the beginning. And so a signal I'm seeing is that these permissions get set early on, they don't get reviewed. And there are two things that come with that. One is that the organization then has to live with what was set at the start without review. And the other one is that sometimes

Underneath the hood, little things change. The permissions don't stay exactly the buttons they were at the start. Some additional capabilities of the tool come in and they hide under one of the buttons that's already being turned on. And so this isn't just an issue of we never looked at it again internally to see whether every team and every function should use the tool in the same way. This is also an issue in regard to how is effectively

A vendor changing the policy of how we work when they change things under the hood. That is my signal

4:13 Sam Rogers

Yeah.

4:14 Markus Bernhardt

for today.

4:14 Sam Rogers

That's fantastic. So we've been kind of sliding towards this situation for a while. With deterministic

4:19 Markus Bernhardt

Yeah.

4:20 Sam Rogers

software, it used to be, you buy the thing, it does the thing, and then there's like a release cycle, a quarterly update, maybe, and we've gone to the rolling update with so many SaaS providers. But what you're describing is like the next

level of that and it's entirely different than managing the software settings or the the permissioning that like stays where you put it.

4:43 Markus Bernhardt

And we probably have non-technical listeners. I I hope we do. And

4:47 Sam Rogers

Yeah.

4:47 Markus Bernhardt

a a good comparison that I like is when a team brought the Microsoft Office suite in, the technical team that brought the suite in used Microsoft Office in the same way that every other team and function in the organization used it.

in this tool there is no difference. Everyone everyone uses Outlook with its permissions in pretty much the same way, and everyone uses Excel definitely in the same way. So we we didn't have to keep an eye on these things. And that that is now drastically changing. And like I said, it's it's a it's a two-fold piece.

we could dive deeper into each of those. One thing that also springs to mind immediately is a technical team brings an HR tool in for recruitment. And when they sign those terms and conditions, and as the agentic piece improves, you've basically s signed a new recruitment policy.

the moment you've agreed to the terms and conditions and the moment they update in the background, they have updated also your your recruitment policy. So for those who think, this is a very technical discussion, I don't care what buttons in the back end of Codex or Claude Code are pressed, for software updates. No, no, no. This is this is very much frontline,

Customer, client facing and internal work. And another example would be privacy notices, right? Different functions and teams have different files moving across their screens and and in their file system. And so we would we would have to watch very carefully where we have which permissions, and they might have to change depending on which team we're working on.

6:14 Sam Rogers

so bringing it to the present situation, how is it that you're working with organizations to even approach this issue, to think about this kind of change and how it is that they're, in standard sense like provisioning software permissions

6:27 Markus Bernhardt

The

Solution is relatively simple and straightforward because we don't need to reinvent the wheel here. We've been in business environments for many decades where things change rapidly. And this is just one of those. We don't just because it's a technical tool, it's just a rapidly changing situation. And it needs decision rights for the people who sign things off. And it needs data contracts and governance for when is it reviewed.

That's basically it. If we can agree that permission settings need to be reviewed on a regular basis, especially when the tool rolls out to a new group of people, that is a good starting point. So we're we're not reinventing the wheel here. This isn't super complicated, but it is a new workflow that has to be implemented. And we need to also ensure that we have the right people in the room.

When we look

7:12 Sam Rogers

Yeah.

7:13 Markus Bernhardt

at these things. So one of the key things is who do we get in the room for that conversation? We don't want 15 people in the room debating on how we do the sign-off and how do we do the reviews, but we do need the right two or three or four people in the room. in my projects, I like having legal there right from the start with a seat. Even if in the first two meetings we're discussing nothing that is particularly quite legal yet.

Involvement is key from my perspective. That teams don't feel like they're brought in either at the last minute or when the house is already on fire. People are part of the decision-making process that leads up to either the smooth running or some friction along the sides. And usually we have friction along the sides, you know, projects don't just run smoothly. And

7:51 Sam Rogers

Yeah. Right.

7:53 Markus Bernhardt

if people have been brought in and we've been inclusive.

then they're very much part of the project team and they're willing to roll their sleeves up and sort out what needs sorting out when we get there. If we bring these teams in last minute or when the house is already on fire, that's that's not the good starting point from a human relationship point of view or from an organizational point of view.

8:12 Sam Rogers

so legal,

how is it that you're finding some of the other appropriate parties for that conversation?

8:17 Markus Bernhardt

Yeah.

It's about looking at the touch points of the tool as it's being rolled out. And

8:21 Sam Rogers

Mm-hmm.

8:21 Markus Bernhardt

every organization has its own products, services, and clients and ways of operating. those would be the key things. let's look at the product, the services, the clients, and the way we're operating, and we will see where in the workflows we have which touch points with which parties. And an internal facing tool.

and that internal team doesn't deal with anything that's particularly secret or might breach privacy boundaries. Those are easy things to handle. So one of one of the obvious examples is, you know, a sales or a marketing team. they don't have very many proprietary.

pieces of information lying around. They do need to have privacy in place for the people that they're marketing to or for the people that they're selling to. But they don't deal with the details of the of the financial accounts and those matters. So those are more of the straightforward example. But if once you go into research and development, into product, into legal, into HR, you have many touch points

That become hotter subjects and then the review needs to involve a wider crowd. Also, tied to the products, but another word I would like to throw in here is sector, right? If you're

9:26 Sam Rogers

Mm-hmm.

9:26 Markus Bernhardt

if you're in medical devices, you're you're in a different sector to if you're in Silicon Valley and you're building social media tech and you can get away with almost anything, unfortunately.

9:34 Sam Rogers

Well, as someone who has worked both in medical device manufacturing and is from the San Francisco Bay Area, has worked with a lot of tech clients, I can say, yeah, their their needs are are definitely overlapping, but very different in in how it is that you approach things there,

9:47 Markus Bernhardt

In in indeed.

9:49 Sam Rogers

as well as you know, different cultures entirely. Well, I'd love to

Lend my signal here as well. It it builds on top of yours. more and more of the AI that we're buying doesn't just like give us the answer. It gives us an answer, but plus a number, 92.7%, you know, some kind of confidence score. somebody has to decide where that line goes. You know, up above this, it's okay, below this, no.

and that line is a default, exactly the kind that we're talking about here. Usually it gets set once by whoever's most comfortable with the tool, like the technical users that you're talking about But everybody downstream inherits that as if it were a fact. and it's not really a fact, it's a policy, as you're describing.

that's treated as a as a static object, everyone else just lives with and assumes that that's the parameters. So some learnings I had twice this week. I was testing a classifier model. I'll come back to that in the subtraction portion, but first, one of its wrong answers came back as 95% confident.

So a rule of only trust the most confident ones would not have caught this. Second, once the results were in, it was very tempting to just move that line. So I had a cutoff that wasn't separating the good from bad very well. moving it three points one way or the other, it would have made every test pass. And the only evidence for

This change would have been that the tests pass, right? So I I actually left it right where it was. So here's a signal. The numbers are arriving faster than the people who should own them. And and finding out who set that cutoff, and figuring out how that conversation happens, how that discovery process happens, is really critical. Because if nobody can

Name the line or like who it is that's setting the line, then it may or may not actually be useful for the goal of that team. something that's set at the organizational level, great. But does it actually help the team get where they want to go? You gave the example of of hiring, you know, and a a talent management workflow.

Hiring actually works differently for different parts of the organization, right? Like it's not even entirely

12:01 Markus Bernhardt

Very much so.

12:02 Sam Rogers

the same process for how you source or vet candidates. it's all supposedly part of the same workflow. But as someone who's worked in talent management for years knows, depending on who you're talking to and what you're sourcing, it's actually quite a different process. so having that

As an inquiry, just with my own direct experience, just in the last week, like I'm doing this myself as one person, and trying to keep that line and make that a conversation piece, make that something that's a dynamic tension in the business, as opposed to a one-and-done fact. is it's very tempting to just lock it down and say this is how it is, but it's not necessarily useful to do that.

12:38 Markus Bernhardt

And

we're often driven by these kind of numbers in business, right? We're used to this. There are there are the sheets, there are the numbers, and if the numbers hit the right spot, we move on. This is a way of thinking, this is a way of operating. especially when it comes to confidence levels and percentages.

there's a lot there to learn and to digest in terms of what is the usability of the number? How did we arrive at the number and what does the number mean? If you're my colleague and I know that you're very unlikely to make a mistake, then then I have a high

13:06 Sam Rogers

Yeah.

13:07 Markus Bernhardt

trust level in you. But that as a colleague would not lead me to saying, well, when he sends a proposal out, no one checks it. We would we would have

We would have checks and balances in place in the organization. We wouldn't be saying, well, if Sam hits 98%, he can just do what he wants. and if he drops naught point one percent below that, we ring the alarm bells and we say everything has changed. No, we've we've we've hardly moved, and you've just changed your take entirely. there is something to be said about the maturity in dealing with percentages and predictions and error rates where

Teams that didn't used to have to think about these things now need to have that level of training and education to be able to interpret these results in the right way. And yeah,

13:48 Sam Rogers

Yeah.

13:48 Markus Bernhardt

it's happening everywhere. I mean, you you can't watch the US Open tennis without it telling you halfway through the second game what IBM Watson thinks who's gonna win the match. And you talk to a lot of people, I I do, about this.

And it's astonishing how many different interpretations people come up with. If IBM Watson, after three games in the first set, says, this person is 62% likely to win. It's incredible how people interpret that number or what they believe, how this number has been arrived at. Especially the amount of people who say, well, this is an intelligent AI system, it will know what it's talking about. Okay.

14:23 Sam Rogers

literally what that means is out of a hundred games, that person would lose 38 of them. And you're trying to figure out which game you're watching. Even to the 95% number, like we talk about a confidence interval or something of 95%,

14:37 Markus Bernhardt

Yep.

14:37 Sam Rogers

that's good enough, 95%. That means out of every 100 decisions that are being made, five of them will be wrong.

that's within acceptable parameters. nothing has gone wrong. That you got exactly the false response, the the non-predictive answer,

14:54 Markus Bernhardt

That you bargained for. Yeah.

14:57 Sam Rogers

five out of every hundred. Well, how many decisions are you making with your AI partner constantly?

How does that calculation even work? Like we know how it works on the machine side. But how does that work on our side? How does that work for us as humans who are interpreting that value and

15:12 Markus Bernhardt

Mm-hmm.

15:13 Sam Rogers

assigning meaning to it? that's a little shakier.

15:16 Markus Bernhardt

it's really interesting that that you brought this one up because this one sort of leads me directly into my subtraction now. Because because

15:22 Sam Rogers

Y Yes, please please go.

15:24 Markus Bernhardt

because my subtraction happens to be the overfocus on a number. And so

My subtraction is

We are now in a very experimental environment and we have to go out and we have to test things. We've we've always done this somewhere in the business, but we're now doing this in every team when we bring a tool in, we have to test and see how it works. And we have to deal with that. And we overfocus on the numbers that we've given ourselves as goals. So one project that I always use as an example is with an insurance company, we look to automate.

A portion, a small portion of the claims under $150. Where the process is relatively straightforward, we think the decisions are relatively simple, and we know that the red flags are all documented. this is where people start when they join that team in the insurance, that's where they start. Small claims. And

16:10 Sam Rogers

Make sense? Yeah.

16:12 Markus Bernhardt

so and so we went into that. There was a time when we sat in a meeting and we said, What do we think we can do here? And

Finger in the wind. We said somewhere between

16:19 Sam Rogers

Yeah.

16:20 Markus Bernhardt

15 and 20% automation would be great after this initial pilot.

16:24 Sam Rogers

Sounds reasonable.

16:26 Markus Bernhardt

well

The interesting thing is first of all that number gets forgotten for a period of time. So we then went in and we looked at the workflows and we documented everything and rewrote an automation tool. And this wasn't an LLM implementation. This was an automation implementation this is like Zapier. If then

16:42 Sam Rogers

Mm-hmm.

16:43 Markus Bernhardt

when this happens, we do this. When this happens we do we do this, and these things need to be checked.

So very, very straightforward, very logical. And we wrote it all up, and then we ran this for three weeks in parallel with the human team. The AI with the automation piece wasn't allowed to do anything. It just ran in parallel and suggested what it would have done had it been doing the job. And then

17:01 Sam Rogers

Mm-hmm. Yeah.

17:03 Markus Bernhardt

after three weeks, we were able to compare what the humans did and what the AI did. And guess what? We found very interesting things about where the AI was unbelievably good and where it fell down. And

We realized that we also saw where the humans were unbelievably good and where the humans were not as consistent as we had always thought they probably were. And

17:21 Sam Rogers

Mm-hmm.

17:22 Markus Bernhardt

so we then consolidated and we learned from that and we improved the automation/slash AI piece. And we ended up being highly confident that we could automate seven to eight percent.

17:34 Sam Rogers

Seven two eight. Yes. Got it. Yeah.

17:36 Markus Bernhardt

Between seven and eight percent. And

that is when the number resurfaces and someone clever will say, But we aim for fifteen to twenty. And and

17:44 Sam Rogers

Yeah. Yeah, you're coming in at the half

l halfway or less from our goal. What a failure. Yeah.

17:49 Markus Bernhardt

Exactly. And

the thing is again, we the the overfocus on one number instead of the overfocus on the actual progress. that is a a really big thing right now. it comes from spreadsheet management. It comes from having numbers on a spreadsheet and comparing them to outcomes and using that without deeper reflection as as a marker for success.

And in an experimental environment, that doesn't work. In an experimental environment where you're trying to find wins and you're willing to have lessons learned, which others might call mistakes along the way, that is simply how it works. And so we celebrated that, and the team celebrated the between 7 and 8% as a huge success. The team was upskilled, the entire team understood automation work much better.

They understood much better what we got wrong in the first round in telling the automation how the decisions are made, which data it needs access to, what flags get raised and when it gets handed off to a human. they are now highly skilled to either attack, you know, the next 10% or the first 10% of a slightly bigger claim, which is a great outcome.

So much has happened. The roles of these people have drastically changed. No one has been let go. everyone is working on the orchestration, and we still need all these people because that guess what? The tacit knowledge that is there, when that rare case comes up, someone on the team will say, that happens once or twice a year. We usually do this, and everyone goes, do we? Okay. the subtraction here is.

We need to stop the overfocus on that number in an experimental environment. when we have a target that's you know, that's not to be taken in the same way as we maybe judge a sales team on the sales target.

19:23 Sam Rogers

Yeah, a target destination versus the success or fail criteria. yeah, that's that's excellent. Well from my subtraction, I'm glad you give the example of this wasn't an LLM kind of implementation. especially with how good the models have been getting lately, I mean just yesterday

OpenAI released a a new one and Anthropic released a new one. I've this week stopped reaching for a large language model by default. Oftentimes that's like the first part of a kind of a triage process of figuring things out. a lot of what we hand these generative models isn't

generative we're not asking it to generate language. And we're not necessarily even using it for reasoning. We're oftentimes using it for sorting. Like which bucket does this go in? Like the the small claims example. You know, is this less than 150 Does this match that? is this about this thing or is it about something else? All of that's classification.

And though there have been tools for this for a long time, they've all been very custom. part of what I've built with with PAICE.work is really just an elaborate classifier of did you catch this kind of error or not is happening behind the scenes. Well, this week I tested a general purpose classifier. It's a small classification model called Jev J E V. It's from TypeSafe.

What I've been testing it against is jobs I would normally give a chat model, usually a small one. So matching plain language questions to a fixed set of research categories. It got 15 out of 15. For keyword search, 10. About a quarter of a second each. So for this show, reviewing 12 of our own episodes.

Cost less than a fifth of a cent. And it's priced at about four cents per million tokens going in, and it's nothing at all for what comes out, which if you've ever had to pay for AI inference, you know is astonishingly low. But this is a subtraction, not an addition. I'm not saying go out and start doing this. I'm

21:22 Markus Bernhardt

Yeah.

21:22 Sam Rogers

saying Jev failed to.

when I asked it whether a legal source actually supported a claim, it missed three out of the 22 that I had given it. And that's that's not really good enough for trusting without a person going to check. it looked like a sort, but it was actually more of a judgment. There was more reasoning involved. So what I subtracted is the default, not the model.

Before anything goes to a general model now, I ask which of the three jobs it is. If it's a rule, it goes to code, it goes to something scripted, the the if this, then that kind of setup that you were referring to. that always beats a confidence score, right? if it's a sort, it goes to a classifier. And if it's a judgment, if it it still needs some reasoning capability, or a person has that, you know, of course.

but the kind of choice I make about where it goes routing that is definitely different this week than it was last week. this is happening, you know, one call at a time one project at a time. I would encourage anyone to take this testing mindset that you're describing, Markus. don't rewire your whole stack in an afternoon.

that's probably

22:33 Markus Bernhardt

Ha ha ha.

22:34 Sam Rogers

not going to go well for you. I know

it's attractive on Twitter and, you know, YouTube you see everybody doing that, but like d y you might want to think about.

22:43 Markus Bernhardt

Yeah. And problem classification is a good thought process at the start of any type of problem we are solving. That that that is not new. Again, we don't need to reinvent the wheel here, right? I love this. because you're you're building on thoughts and processes that have existed for for donkeys years. And so the

23:01 Sam Rogers

Yeah.

23:02 Markus Bernhardt

the beauty here is that.

Sometimes maybe people think, but then I lose something. No, because you can also stagger. Right? In any automation process, in any process. you can have a logical

Step. You can have code do an if then on certain parts of it or on all of it. Whichever. after or before that use a classifier. And use the reasoning capabilities of the old large language model as it's improving, as well. But we always say context is critical.

And context isn't just can I throw more in. Context is also can I throw more really useful stuff in. maybe a Jev output allows a large language model to function much better. And maybe a Python code piece that runs first can get both the Jev as well as the large language model to produce better outcomes.

So really, really love that. I haven't yet had time to play with the Jev model, but

23:59 Sam Rogers

It just came out

just last week. This is like brand new

24:02 Markus Bernhardt

Yeah.

24:03 Sam Rogers

a brand new class of this same technology that we've been working with for a while.

24:05 Markus Bernhardt

Yeah, exactly. but I've

but I've read a little bit about it and I find it unbelievably interesting. So this is something that I've earmarked for either later this week or if if life does what it normally does, then probably for the weekend.

24:17 Sam Rogers

Yeah, I I didn't have any plans this weekend, right? It's AI changed this week. I gotta now now do

24:21 Markus Bernhardt

Yeah, Jev

24:22 Sam Rogers

stuff. Yeah, Jev, the example I come back to as as an analogy with clients is how do you eat an orange? Because everybody has had the experience of eating an orange probably more than one way. Do you slice it first and then peel?

Do you peel first and then eat what's inside? Do you even bother peeling it? Like there's there's not a wrong answer in the sequence here, but they are very different experiences of the same thing that we've all done before: of eating an orange one of these ways. Which is the right way for the team that we're working with? Any which way is fine.

But if we all know what we're doing, then we can do that together much more efficiently there's not like the only way to approach these problems. and just as you're saying, like being able to sequence them in a way we can just test and see which one of these actually works the best for us.

if I'm feeding this to the the five year old, I'm gonna do it this way. If I'm the responsible adult who doesn't want my hands sticky, I might do it this way. You know, like th

25:28 Markus Bernhardt

Exactly.

25:28 Sam Rogers

which way works?

25:29 Markus Bernhardt

Brilliant. I think the world has definitely got more complicated. it used to

be that a clever idea meant we can now go and do the thing.

And these days, a clever idea just means we've given ourselves a lot more homework because we've got some testing to do. We've got some workflow issues that we need to test and evaluate. the better ideas these days put a lot of homework on the table for the team. I used to have the feeling that a really good idea.

saved the team time and made us get there better and quicker. this is something that has definitely changed that honestly is it a surprise that where the where the goods and the ROI is is exactly where hard work was needed.

In order to achieve it, not just with one simple clever idea that anyone else could copy. And that's the competitive edge in the market with all these tools for everyone out there. If we're all using the same tool, then the delta is how we use it and how we're doing our homework around the tool, rather than picking the best tool and then praying, because that's that's what probably a lot of competitors are doing.

26:35 Sam Rogers

the speed to discovery of what the best decision is for a given task in front of us or an entire workflow, like the the speed at which the organization learns and embraces change, like that's the bottleneck that we are

coming up against now. And when there's no shortage of ideas. Lots of ideas. Plenty of ideas.

26:56 Markus Bernhardt

Yeah.

26:57 Sam Rogers

but like we can only do so many at any given time in our human capacity. We we're not gonna test 5,000 at once. We're probably gonna test maybe if we're aggressive, like five different approaches at once. Maybe. probably it's just gonna be one or two. so yeah, yeah, I love that.

I didn't tell you about this, Markus, but after our kind of prep call for this episode, I took some of what we were talking about and made an artifact that is available at sigsub.show/takeaways, the default ledger. So you pick one AI default within your organization, what you're living with, a permission level retention setting.

confidence cutoff on something that somebody sends or approves You write down what it does, who sets it, and whether anybody downstream can change it and what that cadence is. And if you can't fill in the name, you found today's episode in your own office.

I hope that you will go and enjoy it. I will continue to iterate and use it in my own consulting work and my own projects because it's such a useful framing. Thanks to you, Markus, for coming and and talking with us a bit about it.

28:05 Markus Bernhardt

Fantastic. Love that. for those who want a little bit more background on my thinking, you can find me on endeavorintel.com. And once you've had a look at your AI tool that your team uses and you read the defaults and you find out who signed them, you might also want to think about this further and want the longer version.

there are briefs on my website and I've got one called The Decisions Your AI is making without you. So

28:27 Sam Rogers

Mm-hmm.

28:28 Markus Bernhardt

for those who would like to dive deeper into this thought process, you might find that of interest.

28:32 Sam Rogers

Yeah, I I find so much of what's on Endeavor intelligence of interest. maybe you could say just a little bit too about the Endeavour Report and the kind of work that you're doing.

28:43 Markus Bernhardt

I'm an independent research and advisory practice on AI strategy and organizational transformation. I advise enterprise leaders and technology vendors both on their projects. I've published almost a hundred articles on this, and the Endeavor report is sort of my flagship publication. It is a publication on use cases.

it is non-sponsored, non-funded, and not a single vendor gets named. The use cases

29:07 Sam Rogers

Mm-hmm.

29:08 Markus Bernhardt

are

described warts and all. so it's gained a lot of credibility globally.

Because it is told by the people who implemented the use case. Because if it was a paid-to-play.

venture or if it was sponsored, no fortune one hundred company would be allowed to participate in in the research.

29:24 Sam Rogers

Yeah, it's a great way to thread the needle from the organizational perspective, but as someone who's appreciated your work for a while, it's super valuable.

Thanks, Markus.

29:32 Markus Bernhardt

Thanks, Sam.

29:33 Sam Rogers

Great. Well, thanks again, to everyone joining Signals and Subtractions. We live stream every Wednesday, episode out on Fridays, newsletter every Sunday. if you're listening on audio right now, please tap follow.

so that the next one just magically shows up next Friday. Sunday is issue 70 of the newsletter. One signal, one subtraction, one analogy, about five minutes to read. And everything on the show is at sigsub.show. One signal worth watching, one subtraction worth making. Now go find yours.