Best of LinkedIn: Digital Products & Services CW 04/ 05

00:00:00: This episode is provided by Thomas Allgaier and Frennus based on the most relevant LinkedIn posts about digital products and services in calendar weeks O four and O five.

00:00:10: Frennus is a B to B market research company helping enterprises gain the market customer and competitive insights needed to drive customer centric and cost efficient product development.

00:00:20: Welcome back to the deep dive.

00:00:22: We've got a A really interesting stack of sources today, covering late January and early February, twenty twenty six.

00:00:29: Yeah.

00:00:30: And looking through them, I feel like we're, I don't know, maybe turning a corner.

00:00:33: That's exactly it.

00:00:34: It feels like the conversation is finally shifting.

00:00:36: Shifting how?

00:00:37: Because for what, the last two years, it's just been this endless cycle of, look at this cool AI

00:00:42: demo.

00:00:42: Right.

00:00:43: Check out this magic trick.

00:00:44: And the sources from these past couple weeks, they suggest that the shiny new toy hangover is finally clearing up.

00:00:51: Okay, so we're moving from the magic to the mechanics.

00:00:53: The mechanics.

00:00:54: That's the perfect word for it.

00:00:55: It's less about, can AI do this?

00:00:57: And much, much more about, okay, how do we actually build a sustainable operational business with this?

00:01:03: Which I have to say is a relief.

00:01:05: But it also sounds like the hard part.

00:01:07: The plumbing.

00:01:08: It is the plumbing.

00:01:09: Yeah.

00:01:09: But that's where the value is.

00:01:11: We've got insights here on everything from, you know, vibe coding to the real nitty gritty of product ops.

00:01:17: And the common thread seems to be rigor.

00:01:20: Rigor.

00:01:21: Whether it's AI, leadership, or strategy, it seems like the free ride of just winging.

00:01:26: it is over.

00:01:27: Okay, so let's start with that term then.

00:01:29: Vibe coding.

00:01:30: I've seen it all over LinkedIn, all over Twitter.

00:01:33: It sounds like something you'd do in a hammock, not in a serious engineering department.

00:01:37: It does sound super casual, doesn't it?

00:01:39: I mean, in the general sense, vibe coding is this idea of using AI to generate code based on a loose idea of vibe.

00:01:47: You just say, make me a retro landing page and poof, it exists.

00:01:50: Exactly.

00:01:51: And that usually terrifies traditional engineers because the code underneath.

00:01:54: Towards

00:01:54: a mist.

00:01:55: It's a complete mess, right?

00:01:56: It might work, but you do not want to look into the hood.

00:01:59: It's unmaintainable.

00:02:00: Precisely.

00:02:01: But this is where we get our first really big insight.

00:02:03: It's from Lenny Rochetsky's newsletter.

00:02:05: And he was focusing on a workflow by Zivi Arnovitz.

00:02:08: And Arnovitz completely flips this idea on its head.

00:02:11: He does.

00:02:12: He argues that effective AI coding isn't vibey at all.

00:02:15: It's actually rooted in extreme structure.

00:02:19: And this guy, Zivi, he's a non-technical PM, but he's shipping code like a senior dev.

00:02:25: I was reading about his process.

00:02:27: He uses these things he calls slash commands.

00:02:29: It kind of reminded me of old school slack

00:02:31: commands.

00:02:32: It's exactly that.

00:02:33: So instead of treating the LLM like a chat buddy, he treats it like a junior developer who needs very, very specific management.

00:02:39: He has these preset commands like create issue or plan or even peer review.

00:02:44: Let's

00:02:44: pause on that because I think that's a huge shift for anyone listening.

00:02:47: when he types create issue what's actually happening behind the scenes.

00:02:50: So he's not just asking for code.

00:02:52: That command triggers this really rigid template in the AI's instructions.

00:02:57: It forces the AI to state the problem, identify the files, propose a solution before it writes a single line of code.

00:03:05: It stops it from just making something up.

00:03:06: It stops the hallucination.

00:03:08: But the one that really got me was peer review.

00:03:11: That's the killer feature.

00:03:12: Absolutely.

00:03:13: Because when you type peer review, you are forcing the AI to switch personas.

00:03:18: It stops being the builder.

00:03:20: who just wants to get it done.

00:03:21: And it

00:03:22: becomes the critic.

00:03:23: Exactly.

00:03:24: Which is such a smart workaround?

00:03:25: because these models, they're designed to be agreeable, you know, they want to please you.

00:03:30: They rarely say, wait, that's a terrible idea.

00:03:33: So you're basically encoding best practices right into the workflow, giving it a framework to critique its own work.

00:03:40: And here's the best part.

00:03:42: If the AI messes up.

00:03:44: He doesn't just hit regenerate and hope for a better dice roll.

00:03:46: No, he debugs the process.

00:03:48: Yes.

00:03:49: He asks the AI, what in my instructions caused you to make this mistake?

00:03:54: And then he updates the slash command so it doesn't happen again.

00:03:57: That

00:03:57: ties right into this whole slot problem.

00:03:59: There's so much buggy generic AI code out there.

00:04:03: And the inside is the slop isn't an AI problem.

00:04:06: It's a people problem.

00:04:07: Lazy prompts get lazy results.

00:04:09: Zevi actually calls his approach exposure therapy.

00:04:12: Exposure therapy?

00:04:13: Yeah.

00:04:14: He says to start with just a project context doc in chat GPT before you even touch a real coding tool.

00:04:20: You have to train yourself to give the right context first.

00:04:23: It's fascinating.

00:04:24: But you know, on one hand, we have this super structured text heavy workflow that's empowering builders.

00:04:29: On the other hand.

00:04:30: We have a source here, Manuel Silva, who's basically sounding the alarm.

00:04:34: This is such a necessary counterpoint.

00:04:36: He says we're seeing a forty-year regression in design.

00:04:39: Think about it, we spent decades moving away from the command line, you know, MS-DOS typing obscure codes to graphical interfaces, buttons, sliders, visual cues.

00:04:49: And now you open a so-called modern AI app.

00:04:53: And what is it, a blinking cursor in a text box?

00:04:55: Tell me what you want.

00:04:56: Right.

00:04:57: And Silva's point is that we're calling this futuristic, but for an average user, it's a huge cognitive load.

00:05:03: We're making our parents learn prompt engineering just to edit a photo.

00:05:07: It's kind

00:05:07: of lazy design, isn't it?

00:05:08: You're just delegating the work to the user instead of designing a clear button that says remove background.

00:05:13: It is the command line all over again, just with, you know, better autocorrect.

00:05:18: A great interface should show you what's possible.

00:05:21: An empty textbox just assumes you already know.

00:05:23: But wait, if we follow this thread to its logical end, maybe we get to a place where the interface doesn't matter at all because humans aren't even using it.

00:05:32: The no UI future.

00:05:33: Yeah, Akash Gipta had this prediction about agent to agent negotiation.

00:05:37: that kind of blew my mind.

00:05:39: The scenario he describes is so compelling.

00:05:41: Yeah.

00:05:41: Imagine you want a better interest rate on your savings account.

00:05:44: Ugh.

00:05:45: The worst.

00:05:45: I have to log in, navigate, twelve menus, maybe get on the phone.

00:05:49: In the agent to agent future, you just tell your personal AI, get me a better rate.

00:05:53: And your agent pings the bank's agent through an API.

00:05:56: So it's machine to machine.

00:05:58: No humans.

00:05:59: Strictly machine to machine.

00:06:01: They negotiate terms, compare offers, and just report back to you.

00:06:04: I got you four point five percent.

00:06:06: You never touched a screen.

00:06:07: There was no user experience, not visually anyway.

00:06:10: That have massive implications for product managers.

00:06:13: If I'm a PM at that bank, my whole world is button placement and color schemes in this future.

00:06:19: You're invisible.

00:06:20: If your product is just a pretty wrapper on a generic service, you're done.

00:06:25: PMs need to start writing PRD's product requirement documents for APIs.

00:06:31: How does my agent talk to your agent?

00:06:33: What are the negotiation parameters?

00:06:34: It shifts the whole discipline from like psychology of design to economics of APIs.

00:06:40: That makes me wonder about brand loyalty.

00:06:41: If my agent is just getting me the best math, how do you build a brand?

00:06:44: You can't charm an algorithm with a logo.

00:06:46: You can, which is a huge shift.

00:06:49: And it brings us to our second big theme.

00:06:51: the human element.

00:06:52: Because if AI is doing the coding and negotiating, what's left for us?

00:06:56: Judgment, taste.

00:06:58: That's the frontier.

00:06:59: Elaine Barsoom had a brilliant post on this.

00:07:01: She says, scaling with AI is a training problem, but not the way we usually think.

00:07:05: Right,

00:07:05: we're not talking about training the model on GPUs.

00:07:07: No, we have to train the humans.

00:07:10: And she makes this crucial distinction.

00:07:12: You train AI on signals.

00:07:14: Clicks, scrolls, purchases, behavioral data.

00:07:18: But you have to train humans on taste.

00:07:20: which is the judgment of what actually matters.

00:07:22: I love the phrase she used.

00:07:24: Humans retain decision authority and AI keeps input authority.

00:07:28: That's the whole game right there.

00:07:30: The AI gives you the inputs.

00:07:32: Hey, forty percent of users click this, but if you just let the AI make the decision based on that,

00:07:37: you get product bloat.

00:07:38: You get

00:07:39: bloat.

00:07:39: AI doesn't know how to say no.

00:07:42: It just optimizes the metric.

00:07:43: The human has to be the editor.

00:07:45: We have to look at the data and say, yes, users are asking for this, but it ruins the product's simplicity, so we're not doing it.

00:07:51: But taste is such a fuzzy word.

00:07:53: It feels subjective.

00:07:54: How do you pin that down?

00:07:55: Well,

00:07:56: this is where Atira Abdul-Ralph had a great piece distinguishing between product sense and product taste.

00:08:01: OK, this is useful.

00:08:02: Break that down.

00:08:03: So he defines product sense as knowing what to build, solving the right problem.

00:08:08: His example was Slack.

00:08:09: During the pandemic,

00:08:10: right?

00:08:10: Exactly.

00:08:11: Users were feeling isolated.

00:08:12: A purely functional PM or an AI would say, okay, people want to see each other.

00:08:17: Let's add a start zoom button everywhere.

00:08:19: Which

00:08:19: is just more meetings.

00:08:23: The last thing anyone wanted.

00:08:25: Right.

00:08:26: But someone with product sense realizes the problem isn't a lack of meetings.

00:08:30: It's a lack of a hallway.

00:08:32: That spontaneous low friction connection.

00:08:35: And so they build huddles.

00:08:37: That's product sense, so what's product taste?

00:08:39: Taste is about how it feels, the polish, the intuition.

00:08:43: Roth's example here was loom.

00:08:46: Loom noticed users were messing up their recordings and getting frustrated.

00:08:49: A standard solution would be just add a delay button.

00:08:53: Functional, but clunky, kills your flow.

00:08:56: But product taste is realizing the user has performance anxiety.

00:08:59: They're vulnerable.

00:09:01: So Loom adds a quick restart button right next to the stop button.

00:09:03: One click, wipes the take, restarts the countdown.

00:09:06: It

00:09:06: feels like the tool is on your side.

00:09:07: That's haste.

00:09:08: And you can't really prompt an AI to have that kind of empathy.

00:09:11: Not yet, anyway.

00:09:12: Which is why leadership has to change, too.

00:09:15: Roshan Gupta had a post about moving from a product manager to a product leader.

00:09:19: It's a tough transition.

00:09:20: As a PM, your whole value is being the best problem solver in the room.

00:09:24: But as a leader, you have to stop solving the problems yourself.

00:09:27: You have to become a builder of problem solvers, which is terrifying.

00:09:32: You have to let go.

00:09:33: You own the narrative, not the roadmap.

00:09:36: And speaking of narratives, Stephanie Liu had this painfully true post about the illusion of decisiveness.

00:09:42: Oh,

00:09:42: this one hits close to home for so many companies.

00:09:44: It's that moment in the boardroom, everyone agrees on the big bold vision, everyone nods.

00:09:48: The

00:09:48: slide deck looks amazing.

00:09:50: And then Monday morning comes.

00:09:51: And sales is chasing a deal that contradicts the vision.

00:09:55: Engineering is drowning in tech debt, and marketing is off on a tangent.

00:10:00: Everyone is aligned on the vision, but totally misaligned on execution.

00:10:04: And

00:10:04: Lou points out it's the gray dots that kill you.

00:10:06: The

00:10:07: gray dots.

00:10:08: Can we just add this request?

00:10:12: The favor for a big client.

00:10:14: The things that aren't on any roadmap, but they suck up forty percent of the team's energy.

00:10:18: Okay,

00:10:18: so how do we fix that?

00:10:19: This moves us into our third theme, right?

00:10:21: Operations, discovery, the nitty gritty.

00:10:24: The answer seems to be product ops.

00:10:27: It is.

00:10:27: And for a long time, product ops was seen as just, you know, bureaucracy.

00:10:31: More paperwork.

00:10:32: Right.

00:10:33: But Phil Evans makes this great comparison to DevOps.

00:10:36: We don't see DevOps as bureaucracy.

00:10:39: We see it as essential infrastructure for shipping code.

00:10:42: ProductOps is the infrastructure for shipping decisions.

00:10:45: It's a tooling and process that lets the PM focus on strategy instead of fighting fires.

00:10:50: And Ed Biden added some really practical advice here.

00:10:52: He says most leaders have no idea if a team is off track until it's too late.

00:10:56: The big launch day surprise.

00:10:58: The

00:10:58: worst surprise.

00:11:00: So he suggests a four step review rhythm, kickoff, solution review, launch readiness and impact review.

00:11:07: But the one most teams skip is the solution review.

00:11:10: And why is that one so critical?

00:11:12: because it happens before a line of code is written.

00:11:14: It's when the solution is just a sketch, a prototype, that it's the cheapest possible time to fix a misunderstanding.

00:11:20: Before

00:11:21: you have three weeks of engineering sunk cost and emotional attachment to the code.

00:11:25: Exactly.

00:11:26: But all this process can't come at the expense of the customer.

00:11:29: Omar Salem had this great quote, a product idea without validation is just a well-packaged guess.

00:11:35: I love that.

00:11:35: Validation isn't a phase you check off, it's a principle.

00:11:38: You have to challenge your assumptions.

00:11:40: not just ask users what they

00:11:42: want.

00:11:42: And speaking of challenging assumptions, Michael Goitin shared the strategy test called the toggle method.

00:11:47: I thought this was a brilliant mental model.

00:11:50: It's a tough one.

00:11:51: Yeah.

00:11:52: So every strategy have a where to play your customers and how to win your differentiator.

00:11:57: Okay.

00:11:58: Goitin says to take your how to win and toggle your where to play.

00:12:03: to a competitor's customer base.

00:12:05: Let's try it.

00:12:06: Say my where to play is creative agencies and my how to win is a simple intuitive interface.

00:12:13: Okay, now toggle the customer.

00:12:14: Let's say accountants.

00:12:16: Does simple intuitive interface work as a differentiator for accountants?

00:12:20: Well,

00:12:21: yes.

00:12:22: Everyone wants a simple interface.

00:12:23: Exactly.

00:12:24: Your strategy failed the test.

00:12:26: If your differentiator works just as well for a totally different segment, it's not a differentiator, it's just table sticks.

00:12:32: So a real strategy would be something like, we visualize timelines in a way that only makes sense for video production workflows.

00:12:38: Yes.

00:12:39: Because if you showed that to an accountant, they would hate it.

00:12:42: That means you've made a real choice.

00:12:43: You've sacrificed one market to win another.

00:12:46: I feel like a lot of PowerPoint strategies would fail that test.

00:12:49: Most

00:12:49: would.

00:12:50: Okay.

00:12:50: Before we wrap up, I want to get to that technical nugget you found.

00:12:53: This is for the e-commerce folks listening.

00:12:54: It's from John Roberts about how LLMs like ChatGPT actually see products on Shopify.

00:13:00: Right.

00:13:01: This is a perfect example of mechanics over magic.

00:13:04: We just assume AI knows things.

00:13:06: But Roberts explains that these LLMs, they often prioritize reading the body thermal of a product page.

00:13:13: And they do not necessarily read the SEO meta descriptions we've all spent years optimizing.

00:13:18: That's

00:13:18: huge, right?

00:13:19: The meta description is often just marketing fluff.

00:13:21: The body terminal is where the rich details are, materials, the use cases, the siding.

00:13:25: That is the data the LLM needs to answer a real question.

00:13:28: So

00:13:28: the advice is, stop writing for a search engine spider from twenty fifteen.

00:13:33: Write detailed, rich descriptions in the main body text.

00:13:37: If you want an AI to recommend your shirt when someone asks, what's a good shirt for a hot climate?

00:13:43: The answer, one hundred percent organic, cotton, breathable weave, needs to be in that body text.

00:13:50: If it's buried in a metatag, the AI might just miss it.

00:13:53: It all comes full circle back to the beginning.

00:13:55: Whether it's ZV's slash commands or writing product descriptions, you have to understand how the system actually works, not just how you wish it did.

00:14:04: The era of vibes is ending.

00:14:06: The era of understanding the system is here.

00:14:08: So if you're listening, I guess the question to ask yourself is, Are you building for the shiny toy version of the world?

00:14:14: Or are you building for the reality of agents, rigor, and structured workflows?

00:14:18: And maybe one final thought to leave with.

00:14:19: We talked about taste being this uniquely human skill.

00:14:22: But as we get better at encoding our preferences into these systems, into these slash commands, are we just teaching the AI to have taste?

00:14:29: At what point does the human element just become another dataset we hand over?

00:14:33: That is a slightly terrifying thought to end on.

00:14:36: but a necessary one.

00:14:38: If you enjoyed this episode, new episodes drop every two weeks.

00:14:41: Also check out our other editions on ICT and tech, artificial intelligence, cloud, sustainability, and green ICT, defense tech, and health

00:14:49: tech.

00:14:50: Thanks for listening.

00:14:51: Don't forget to subscribe for more insights from the next calendar week.

00:14:53: See you then.