Best of LinkedIn: Digital Products & Services CW 32/ 33

00:00:00: This episode is provided by Thomas Allgaier and Frennis, based on the most relevant LinkedIn posts about digital products and services from CW-Thirty Two and Thirty Three.

00:00:09: Frennes is a BDB market research company that supports enterprise product teams with building feature by feature competitive intelligence That shows exactly how their product stacks up against the competition.

00:00:20: You can find more info in the description.

00:00:22: So imagine logging into work tomorrow Right.

00:00:25: And you realize your primary software, the user interface it's just completely gone.

00:00:30: Oh man yeah.

00:00:31: Dashboards no uh...no Dutton's Just a blank text box where an agent is essentially waiting for your intent.

00:00:39: It

00:00:39: sounds like science fiction but its actually happening right now.

00:00:42: Exactly

00:00:43: So today we are doing deep dive into why the traditional software application Is kind of dying And why the professionals who build them are facing this massive existential crisis.

00:00:52: Yeah, we're unpacking top digital products and services trends across industry.

00:00:57: looking at sources from CW-thirty two to thirty three Undeniable the foundational unit of product development, you know The product manager is being forced into this complete reevaluation Of what their job actually?

00:01:10: Is because AI is rewiring not just how we build but the very definition of What a product even

00:01:17: it's moving so fast right like that.

00:01:19: The technology are moving So fast at the actual job title is already changing.

00:01:23: yeah there was this post we saw from Akash Gupta, and he highlighted these recent comments from Satya Adela about a major shift happening over at LinkedIn.

00:01:32: Oh right the builder thing?

00:01:34: Yeah

00:01:34: exactly so.

00:01:35: they are moving away from an associate product manager program and moving toward an Associate Product Builder Program.

00:01:40: Right!

00:01:41: And builder is such.

00:01:42: you know it's deliberate departure for manager.

00:01:44: Huge difference

00:01:45: yeah...and Nadella specifically named evaluations or evils as they call them As The Core New Skill that these product builders have to own.

00:01:54: So the job is no longer about like managing a timeline or, you know writing those endless requirement documents.

00:02:01: The worst

00:02:02: right?

00:02:02: The job has now evaluating the outputs of AI models within a very specific process.

00:02:07: in That framing it really brings to mind that debate sparked by Alan Samuels.

00:02:14: He referenced this recent CPO Insights report that was forecasting that product managers would be completely obsolete by twenty thirty.

00:02:22: Wow, obsolete completely gone.

00:02:25: but he pushed back on That with his really great counter argument?

00:02:29: He compared AI's effect on product management to to what Excel did to accountants.

00:02:35: Oh, that's a perfect

00:02:36: analogy!

00:02:36: Right like...Excel didn't eliminate accounts it just removed the manual arithmetic It

00:02:40: removed the drudgery and-and just left the judgment because I mean Excel crunched the numbers sure but the accountant still needed to actually look at the spreadsheet in flag of structural error Exactly.

00:02:51: And so AI handles the drafting The summarizing you know basic user research synthesis and That leaves the PM to exercise Just pure product judgement.

00:03:00: Okay, but let's unpack this for a second.

00:03:02: Because if AI is doing all the grunt work—the drafting and early research—how do new PMs actually learn that judgment in first place?

00:03:09: That is the million-dollar question!

00:03:11: Right because historically grinding through user interviews and writing terrible initial stacks like...that is exactly how a junior PM built their product sense.

00:03:22: Yeah…and tension is basically central paradox of today.

00:03:27: Lesia Magas touched on this beautifully in The Sources.

00:03:30: She observed that new PMs can prompt AI to do almost anything today, except build the actual underlying judgment.

00:03:46: But you still need fundamental product sense to know if that feature actually solves the user's root problem.

00:03:53: So reading those foundational theories on decision-making and customer psychology, it is more critical now because the AI will confidently serve this elegant solution for completely wrong problems.

00:04:03: And If You Don't Know What A Good Outcome Looks Like, you'll just accept whatever system spits out

00:04:08: Which Is Completely Changing How Companies Hire Right Now.

00:04:11: Mo Ali pointed out that landing an AIPM job today requires answering these really complex scenario questions.

00:04:17: Oh, like what?

00:04:18: Well interview loops are moving away from the standard prioritization frameworks.

00:04:22: Candidates are now asked to explain things like model routing economics.

00:04:27: Wait!

00:04:27: Model Routing Economics.

00:04:28: I was just talking about server cost.

00:04:30: Let's get a quick ELI-V on.

00:04:32: that actually means for PM day today.

00:04:34: Okay so it is essentially profit and loss management of API calls.

00:04:38: You have decide which underlying language models handles specific user request within your app.

00:04:45: So a highly complex reasoning task might need the most expensive heavy-duty model.

00:04:50: Right, The big ones!

00:04:51: Yeah

00:04:52: but a simple text classification tasks.

00:04:55: you can route that to cheaper faster models.

00:04:57: so PM has constantly balanced cost of compute against value for users action.

00:05:03: Wow, so they're functioning almost like an algorithmic traffic cop.

00:05:06: A traffic cop who also has to deal with constant accidents.

00:05:09: because Ollie also noted that interviewers want to know how you handle a model's ninety percent accuracy rate?

00:05:16: Right

00:05:16: what happens in the ten-percent of cases where it just

00:05:18: fails?

00:05:20: are those failures even detectable by the system?

00:05:23: Are they reversible for the user?

00:05:25: How do you recover the user experience if an autonomous agent fails halfway through a multi step workflow?

00:05:30: That is so much responsibility and companies are clearly willing to pay a massive premium for that specific judgment.

00:05:38: I mean, we saw that job opening at Baton For an AI product manager And the salary range went all the way up to three hundred

00:05:46: thirty thousand

00:05:46: dollars.

00:05:47: It's wild!

00:05:49: Entirely new sub skills emerge to support this.

00:05:52: Yeah, Amy Mitchell mentioned his concept of the knowledge DJ.

00:05:56: I love that term The Knowledge DJ.

00:05:57: it is a perfect descriptor for the modern pms daily reality.

00:06:01: It's basically the active curation Of what context belongs in a products AI system?

00:06:06: yeah And it makes sense mechanically right.

00:06:07: the PMS job Is to retire stale material.

00:06:10: so it doesn't dilute the AI systems.

00:06:12: understanding like if you just dump every outdated document Every deprecated feature spec that old slack thread from three years ago into your AI's context window, the model just gets confused.

00:06:24: Oh it completely loses the

00:06:25: plot!

00:06:26: Right?

00:06:26: It starts hallucinating instructions based on how this software used to work.

00:06:30: so the PM has to actively DJ the knowledge base deciding what the AI is actually allowed to remember.

00:06:37: but here's the catch with that.

00:06:39: curating that knowledge base creates a totally new bottleneck.

00:06:43: How so?

00:06:44: Well,

00:06:44: if the AI is instantly digesting that curated context and generating code or specs in seconds The pressure in the overall system doesn't just disappear.

00:06:53: It just moves downstream.

00:06:54: Ah I see it moved to figuring out what to actually build in the first place.

00:06:58: Exactly Let's call this the velocity gap because the execution phase Is on hyperdrive.

00:07:03: Yeah And that leaves the discovery phase Just completely overwhelmed.

00:07:06: Mal Schultz framed This phenomenon brilliantly.

00:07:09: He observed product managers spinning up tools like Claude Code or Figma Make and they'll just prototype for hours.

00:07:17: They generate dozens of screens, they refine the UI details and move at incredible speed.

00:07:22: And do this because it provides a massive psychological dopamine hit.

00:07:26: It feels work

00:07:27: Right!

00:07:28: Makes them feel incredibly productive.

00:07:29: But are spending zero time defining actual problem that we're trying to solve?

00:07:35: Yes Prototyping feels like progress while stopping to talk to a customer just feels like stalling.

00:07:42: So what does this actually mean for the business?

00:07:44: Are we just building the wrong things faster,

00:07:47: I mean that is The Grim reality.

00:07:49: Jeff Goff health noted that AI hands product teams several extra hours a week by speeding up production.

00:07:54: right but in most organizations That time goes straight back into the backlog.

00:07:58: Teams Just use the Time To Ship More Output Against Unvalidated Assumptions.

00:08:03: Wow So they're amplifying the shipping part of feedback loop, but completely ignoring learning.

00:08:08: Exactly!

00:08:08: They are feeding the feature factory beast instead protecting that newfound time for true discovery.

00:08:14: But research teams at least try to keep up with this velocity Right.

00:08:19: Ryan Glasgow pointed out that researchers are facing this five-axe buildings bead by creating multiagent research workflows, right?

00:08:27: They're using the model context protocol or MCP to connect their tools and automate the tactical parts of research.

00:08:33: Yeah, MCP is everywhere right now

00:08:35: it really is.

00:08:36: we see mcp mentioned constantly in these sources but for those of us who aren't engineers how does What is technically happening in the background that makes these multi-agent workflows so fast?

00:08:49: Okay, think of how software used to integrate.

00:08:51: A human had to open a browser click a tab export a CSV file from a database and then upload it into an analysis tool.

00:08:59: So tedious Right.

00:09:00: MCP eliminates the human interface entirely.

00:09:03: It's an OpenStandard that allows AI models securely connect directly with your company's data sources.

00:09:08: The agent queries the database reads the user-interviewed transcripts and synthesizes the findings in the background without ever needing a screen.

00:09:16: So it's kind of like a sous chef having direct access to the pantry, right?

00:09:20: And prepping the ingredients before the head chef even walks into the kitchen.

00:09:23: That is a great way to visualize that they are automating the plumbing research.

00:09:27: so maintain their strategic

00:09:30: craft.

00:09:30: Right

00:09:31: But even with faster research, there is this massive disconnect in how leadership views the building speed.

00:09:38: Johnny McDonald ran a survey exposing major reality check here.

00:09:43: Ninety-two percent of sauce leaders surveyed said they feel satisfied with on boarding visibility.

00:09:50: They believe know exactly what's happening.

00:09:54: But wait, when you look at the actual instrumentation only twenty four percent actually track where users drop off.

00:10:00: So they feel confident but they lack the actual data.

00:10:04: Why?

00:10:05: The delusion like what is a psychology behind that?

00:10:08: McDonald calls it confidence trap?

00:10:10: the sheer velocity of AI driven output creates this psychological illusion of progress.

00:10:15: Shipping ten features week feels like your solving the user's problem.

00:10:19: because you're busy

00:10:20: right An outcome metrics tell you a user eventually churned, but because you are moving so fast.

00:10:25: You never build the tracking to tell why they turned.

00:10:28: your sense of control outpaces The actual instrumentation backing it up

00:10:31: which forces A very uncomfortable question.

00:10:34: honestly if we know We're building faster than we can validate and we know we Are falling into this confidence trap?

00:10:41: Why aren't organizations fixing the system?

00:10:43: well I mean, the root cause isn't the AI technology.

00:10:47: It's legacy organizational structures.

00:10:49: Yeah!

00:10:49: The operating models are where friction truly lives.

00:10:53: Axel Fornier referenced a KPMG analysis finding that fewer than one in five digital transformations actually deliver the promised value.

00:11:00: Only

00:11:00: One In Five

00:11:01: And diagnosis always comes back to same structural flaw.

00:11:05: These operating models were built for planning not learning.

00:11:09: They're designed execute pre-determined twelve month roadmap not to adapt new information gathered by an AI agent on a random Tuesday.

00:11:17: And Joy Adamson shared the perfect example of how this plays out in the ground.

00:11:22: She noted that forty percent of B-to-B product teams cite poor prioritization as massive problems.

00:11:29: But when you look closely at mechanics, it isn't a product problem It is governance issue.

00:11:35: Exactly A team will use structured mathematical framework like Rice in a reach impact confidence effort to objectively score and prioritize features.

00:11:45: Yeah, they do all the deep analytical work right.

00:11:48: And then executive leadership steps in and simply overrides the scores because of a knee-jerk reaction to a competitor's press release or just a desperate short term revenue push.

00:11:57: Right, if you are sitting on a product team right now and your VP overrids your rice scores on a whim You were living this reality.

00:12:05: Oh yeah.

00:12:06: And companies keep going through these massive expensive transformations.

00:12:09: They renamed teams into product teams and they hired people with the title Product Manager.

00:12:14: But If The Authority To Say No doesn't actually change, nothing is different.

00:12:19: Steve Bazanet and Manus Deb call this the product theater trap.

00:12:22: Product Theater Trap?

00:12:23: I like that!

00:12:25: Yeah it occurs when an organization adopts the product titles in the agile rituals but the actual operating capability stays completely flat.

00:12:34: so The Orc Chart changes sure But the funding mechanisms ,the governance structures And the success metrics do not.

00:12:40: And speaking of keeping up with massive organizational shifts and, you know cutting through the theater.

00:12:44: This is a great time to hit subscribe on whatever app your using so that you don't miss future deep dives.

00:12:51: We are tracking these changes every week

00:12:53: Absolutely.

00:12:54: So how do we actually measure if an organization has escaped that product theatre?

00:12:59: Well Richard Sonnenblik suggests a brilliant lit mist test for this To see If AI Has Actually Transformed An Operating Model.

00:13:06: You Just Ask Three Sharp Questions.

00:13:08: Okay Lay them On Me

00:13:09: One Has it eliminated a meeting?

00:13:12: Two, has it shortened an approval chain?

00:13:15: and three...has changed the funding decision.

00:13:18: Wow!

00:13:18: That's so simple but so effective.

00:13:21: because if they answer to all three is no you haven't transformed anything.

00:13:24: You've just bought an expensive stopwatch for the exact same rate.

00:13:27: exactly your producing these acts in reports Just faster

00:13:31: right?

00:13:31: you gained an isolated efficiency But that operating model remains a bottleneck.

00:13:36: So we have PM roles evolving to focus on evaluation and model routing.

00:13:41: We have a massive velocity gap between building and learning, operating models struggling with the pace...

00:13:47: It's

00:13:49: a lot!

00:13:49: And all of these shifts are ultimately culminating in totally different type end product.

00:13:55: The operating model isn't the only thing changing.

00:13:57: The product itself is kind of vanishing into the background.

00:14:00: We are definitely entering the era of the agent-driven or headless

00:14:04: application.

00:14:05: Headless application

00:14:06: Yeah.

00:14:07: Casey Hill laid out how the old sauce playbook operated for the last fifteen years.

00:14:11: It was entirely focused on DAU, or daily active users.

00:14:15: The goal was to build a sticky product force users into your specific interface and built A Daily Habit there.

00:14:21: Right!

00:14:21: The visual interface with the company's defensive mode

00:14:24: Exactly.

00:14:24: But AI completely breaks that assumption.

00:14:27: With tools like Clay or Zapier today, users are increasingly accessing them without ever visiting their websites.

00:14:33: They just use Claude or ChatGPT which connects to those services through the model context protocol we discussed earlier.

00:14:40: EMCP.

00:14:40: Yes

00:14:41: So The language model is the interface and the sauce product Is a headless engine running silently in the background.

00:14:47: Wait but it sounds like restaurant trying To keep customers out of its dining room.

00:14:51: Yeah, a little bit.

00:14:52: Like if the new AI product strategy is to actively get people not to log into your product.

00:14:58: how does a company actually make money or defend its territory?

00:15:02: If nobody visits The Interface what stops a competitor from just replacing you?

00:15:07: well...the defensive mode shifts away from interface engagement and moves entirely toward proprietary data and distribution.

00:15:14: okay If a company's only value was a slick UI and clever dashboard, they are incredibly vulnerable right now.

00:15:20: But if their value is unique dataset or highly reliable complex workflow engine AI actually makes that mode stronger.

00:15:29: Agents can access it instantly making the software utility layer

00:15:32: That make sense.

00:15:33: It changes how we evaluate software purchases.

00:15:35: entirely doesn't?

00:15:36: Oh, totally.

00:15:37: Johnny Longdon showcased a rebuilt vendor comparison tool for A-B testing platforms in the sources and instead of ranking these platforms on which one advertised the most flashy AI features in their user interface he ranked them whether an agent could actually operate.

00:15:54: Wait, really?

00:15:55: Yes.

00:15:56: He tested whether an autonomous agent could create a test launch it and read the results through MCP without human ever clicking thru platform screens.

00:16:06: That

00:16:06: is wild!

00:16:09: Software must be readable and operable by machines first, and humans second.

00:16:13: Which

00:16:13: creates a completely wild design challenge.

00:16:15: Yeah.

00:16:16: Marcelo Dentus pointed out that generative AI is moving us from command-based interfaces to intent based interfaces And the mechanics

00:16:23: of that shift are profound.

00:16:25: I mean in a command base system The user carries cognitive burden.

00:16:28: knowing sequence steps

00:16:30: Right Click left menu Select audience drop down Input budget hit submit.

00:16:34: You have know them app

00:16:35: Exactly, but in an intent-based system the user just states the desired outcome.

00:16:40: You'd say optimize my ad spend for upcoming holiday weekend.

00:16:44: That's it?

00:16:45: The system then determines necessary procedure to achieve that goal.

00:16:49: So design challenge is no longer about discoverability like making sure the user can find a right button.

00:16:54: It is about capturing intent accurately and managing authority.

00:16:59: Managing Authority Let us dig into that.

00:17:01: Authority is the critical frontier of product design.

00:17:04: now A system might perfectly understand your intent to optimize ad spend.

00:17:09: It might know the exact mathematical procedure to execute it, but the design question is does that have authority increase budget by twenty percent on its own?

00:17:17: Oh right!

00:17:18: Does this has an underperforming campaign without asking?

00:17:22: So designing permission guardrails for when a systems allowed act versus when must ask human approval that replaces traditional UI designs.

00:17:32: You weren't designing buttons anymore, you were designing trust boundaries.

00:17:35: It's a completely different paradigm and if we pull all of these threads together it raises really provocative question about the future professional work.

00:17:44: Think about it.

00:17:45: The cognitive burden on navigating software and constructing workflows is moving from human to AI.

00:17:51: Our applications are becoming increasingly headless and agent-driven.

00:17:55: If this continues, will the next generation of professionals even know how the underlying processes in their own businesses actually work?

00:18:03: That's a scary thought!

00:18:04: Right... Or they simply state an intent to entirely rely on AI's authority for the job?

00:18:10: Yeah that is something you want AI and agentic systems, sustainability in green ICT, defense tech and health tech.