Best of LinkedIn: Digital Products & Services CW 42/ 43

00:00:00: This episode is provided by Thomas Allgaier and Frennus.

00:00:03: Based on the most relevant LinkedIn posts about digital products in services from CW-Forty Two and Forty Three, Frenness is a B to B 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.

00:00:21: You can find more info in the description

00:00:23: And looking at this highly curated stack of insights from across LinkedIn, our mission today is to just cut through all that noise.

00:00:30: We want to identify the absolute top trends in digital products and services right now because if you are a digital transformation professional You are definitely already feeling the ground move under your daily workflows.

00:00:42: So today we're focusing strictly on what's actually working in the field.

00:00:46: according to leaders who live

00:00:48: it Right What have I told you?

00:00:50: giving developers an AI tool code ten times faster, might actually be the exact thing that destroys your product timeline.

00:00:58: Oh man!

00:00:59: That is a wild thought but it's so true.

00:01:02: Welcome to The Deep Dive everyone.

00:01:03: let's really trace that logic because the first massive shift jumping out of these sources about where the friction Ashley lives in modern product development today

00:01:13: Yeah...the bottleneck has totally moved

00:01:15: Right.

00:01:15: for longest time.

00:01:16: primary constraint was always engineering capacity.

00:01:20: Well, we didn't have the developer hours.

00:01:22: You could never hire enough engineers

00:01:23: exactly.

00:01:25: But now with AI coding agents entering the chat that constraint is just vanishing.

00:01:31: but The bottleneck didn't disappear it Just it moved down the assembly line entirely to decision quality.

00:01:37: It's a classic shifting bottleneck really right?

00:01:40: and John Leslie brought up A brilliant point regarding queuing theory To explain the mechanics of this.

00:01:45: oh queueing theory.

00:01:46: I love this stuff Right?

00:01:48: its been governing work systems for like over a century.

00:01:50: And Leslie points out that even with AI-assisted coding making development incredibly fast, little's law still fundamentally applies.

00:01:59: Wait remind me how Little's Law goes again?

00:02:01: So average lead time equals your work in progress divided by your throughput because AI lets you start way more work than ever before.

00:02:09: if You start more then you can finish.

00:02:11: Your Lead Time actually grows.

00:02:13: Okay let's unpack this Break it down for Me.

00:02:15: How does That Actually Look In Production When We Push The System To the Limit?

00:02:19: Well, Leslie highlights the math and it exposes a massive operational blind spot at say eighty percent utilization of your system's capacity.

00:02:28: You get a weight multiplier of four okay but And here is the crazy part if you push that to ninety five percent utilization That weight multiplier spikes to nineteen.

00:02:39: Wait I want pause on that.

00:02:40: Why does a fifteen percent increase in utilization cause the wait time to multiply by nearly five?

00:02:48: It's exponential.

00:02:50: Is it because at ninety-five percent, there is actually no slack left in the system to handle any variants?

00:02:55: Like if one code review takes an extra hour and a whole line just backs up...

00:02:59: That is the exact mechanism yeah.

00:03:02: Slack is what absorbs variance in any complex system.

00:03:05: right you need that buffer.

00:03:06: exactly

00:03:07: when AI allows developers to just churn out pull requests at light speed The System hits ninety five percent utilization almost instantly.

00:03:15: oh wow.

00:03:16: so Making the coding phase ten times faster doesn't actually mean that product ships ten time faster.

00:03:22: It just takes a massive pile of work and slams it into next stages.

00:03:27: Which would be code review, integration...

00:03:30: Yeah!

00:03:30: And critically human-product decisions.

00:03:33: Okay.

00:03:33: so its essentially like expanding highway to ten lanes but keeping single one lane toll booth at exit.

00:03:39: That's perfect way.

00:03:40: look at it.

00:03:41: We aren't increasing flow traffic.

00:03:44: We're just generating a highly efficient, massive traffic jam right in front of the toll booth.

00:03:49: And the pressure on people operating that toll booth becomes immense

00:03:52: Which creates very dangerous context gap.

00:03:56: I mean Constantine Richter shared an anecdote That perfectly illustrates what happens when The highway meets the tollbooth.

00:04:02: What did he say?

00:04:03: He talked about building A product roadmap using Claude In just thirty minutes.

00:04:07: Thirty

00:04:07: minutes Wow.

00:04:08: And he admitted that doing it so fast made him incredibly impatient with his actual team.

00:04:14: He's sitting there thinking, I just drafted this in half an hour.

00:04:16: why are you telling me the deployment will take weeks?

00:04:19: Because reality doesn't compile in a vacuum.

00:04:23: Exactly.

00:04:24: Richard noted that a seemingly trivial one line change on that AI generator roadmap could quietly touch a client settlement flow and alive banking system.

00:04:34: Yes The stakes were high.

00:04:36: The engineers and product managers, they know what a change actually touches.

00:04:40: And crucially What could break alive ledger at two AM?

00:04:43: AI makes the drafting process effectively free but it absolutely does not lower the cost of being wrong in production.

00:04:50: No Not At All!

00:04:51: The financial penalty of Alive Banking Error is just as high As It Was Before LLMs Even Existed.

00:04:57: And Honestly The Polish Of AI Actually Masks That Danger.

00:05:01: Jeff Goff-Elf made an observation that pairs perfectly with Richter's story.

00:05:05: What does this take?

00:05:06: Goff Elf describes AI output as being dangerously convincing.

00:05:10: Oh, that is a great phrase!

00:05:12: Dangerously convincing.

00:05:14: It looks like a final design, it reads like solid well-researched market analysis complete with you know confident formatting.

00:05:21: That's got the bullet points and the bold text

00:05:22: Exactly.

00:05:23: But he warns that this just accelerates our ability to generate guesses without reducing the underlying uncertainty of whether those guesses are tied to reality.

00:05:32: We have treat AI output as high fidelity assumption.

00:05:35: Yes

00:05:36: exactly High Fidelity Assumptions.

00:05:38: Alexander Hipp builds on this beautifully by pointing out solving the wrong problem efficiently is Totally pointless.

00:05:44: It's

00:05:44: completely pointless

00:05:45: In this environment, AI just allows enterprise teams to scale a bad idea twenty times faster.

00:05:50: So if AI lets us scale bad ideas at twenty x speed How do we stop those bad ideas before they clog up the engineering toll booth?

00:05:58: Right what is the filter?

00:06:00: The only reliable filter left as strategy.

00:06:03: But... A strategy that cannot say no isn't a strategy it's a corporate wish list.

00:06:09: I actually found a great diagnostic for that wish list.

00:06:12: Akash Gupta shared an insight on Tim Herbig's test for product strategy?

00:06:16: Oh yeah, the strategy tests!

00:06:18: Yeah specifically highlighting The Flip Test created by Roger Martin.

00:06:21: have you seen this?

00:06:22: remind me how the flip test works.

00:06:24: so You take your stated strategic choice and you state the exact opposite of it.

00:06:29: If the opposite is plainly absurd then your original statement was just a platitude.

00:06:34: okay give Me An example.

00:06:35: So if Your Strategy Is um provide a great customer experience.

00:06:40: The Flip is provided.

00:06:42: terrible customer experience,

00:06:43: right?

00:06:44: Which no one would ever choose

00:06:45: exactly.

00:06:46: No enterprise whatever choose that.

00:06:47: so it's not a real strategy.

00:06:49: but if you say target repeat buyers of luxury cars in Germany the flip is Target first time buyers of budget cars across Europe

00:06:58: Ah I see.

00:06:59: That opposite actually makes business sense for competitor.

00:07:02: therefore your original statement Is A deliberate choice with Real trade-offs.

00:07:06: It forces a trade-off, which is incredibly clarifying.

00:07:09: But once you have that deliberate strategy... ...you have to measure it with outcome discipline.

00:07:14: Outcome Discipline is tough though!

00:07:16: It really is.

00:07:17: Tom and Olaway cited the twenty-twenty six OKR benchmark report And turns out fifty two percent of key results currently measure activity rather than actual outcomes.

00:07:28: Fifty Two Percent.

00:07:30: But I mean, that makes sense because activity is easy to measure when AI's inflating your throughput.

00:07:34: Exactly!

00:07:35: You can just point it at fifty shipped features and claim victory even if churn hasn't dropped a single percentage point.

00:07:41: but That illusion is colliding with the pretty major structural shift in leadership right now.

00:07:46: Alex Bass points out that seventy percent of chief product officers Now hold direct P&L responsibility.

00:07:52: Oh wow they own The Profit and Lost now.

00:07:54: Yeah So when the CPO owns the P&L, shipping fifty features is completely meaningless if they don't drive top-line revenue or reduce operational costs.

00:08:03: Right Bass argues this makes willingness to pay research infinitely more critical than development cycle time.

00:08:10: You have to know a client will actually pay for that feature before you burn developer cycles on it However fast those cycles are

00:08:16: That tracks.

00:08:17: But here's where he gets really interesting.

00:08:20: I'd argue there's massive contradiction happening in corporate behavior right now.

00:08:24: How so?

00:08:25: McKinsey Hughes highlights that despite this new P&L accountability, product teams will still routinely disappear into a cave for like six weeks to create beautiful comprehensive forty-slide strategy decks.

00:08:37: Oh I've

00:08:37: seen those decks!

00:08:38: Right?

00:08:39: But they do it without ever picking up the phone to talk with single churned customer.

00:08:42: They are building artifacts instead of building understanding.

00:08:46: Exactly Hughes asks aren't these decks just incredibly well formatted assumptions?

00:08:51: Yeah If they haven't been pressure tested against brutal market feedback, They aren't a strategy.

00:08:57: They are just a presentation of what the executive team hopes is true

00:09:01: and it's hoping for the best

00:09:02: Absolutely.

00:09:03: And by the way if you the listener Are enjoying this rapid fire breakdown into mechanisms driving tech right now make sure to subscribe To The Feed so catch all future editions Of Deep Dive.

00:09:13: Definitely Subscribe.

00:09:15: So if strategies frequently being masked By beautiful decks and AI Is churning out actual code.

00:09:21: Who is driving the day-to-day vetting?

00:09:24: That's a million dollar question.

00:09:25: Like,

00:09:26: what is actually happening to product manager role right now?

00:09:29: The ground has shifted completely.

00:09:31: Moa Ali shared metric from recent meeting with Fortune five hundred VPs.

00:09:35: What was that metric?

00:09:36: Roughly seventy percent of senior leaders have stopped hiring associate or junior product managers.

00:09:42: Seventy

00:09:43: percent?!

00:09:43: That an entire entry level tier just being wiped out.

00:09:47: We are essentially burning down training grounds.

00:09:50: The expectation from leadership has fundamentally shifted.

00:09:53: They now expect their existing senior PMs to build AI agents, To absorb the grunt work that junior PMS used

00:10:00: to handle.

00:10:01: Like what kind of gruntwork?

00:10:02: You know initial competitive research synthesizing customer feedback calls basic reporting.

00:10:08: Alley notes That the line between product and engineering is completely blurring because pms are Now architecting these internal tools themselves.

00:10:16: I mean, i can see the optimistic angle here.

00:10:18: Right?

00:10:18: Brian Sargent framed it perfectly when he said AI didn't kill product management.

00:10:23: It fired The paperwork

00:10:25: Fired the paperwork!

00:10:26: I like

00:10:27: that.

00:10:27: Yeah...it removes the administrative layer-the endless backlog grooming and theoretically leaves PMs to focus entirely on customer focused decision making.

00:10:35: Theoretically yes

00:10:37: But..I have to push back hard On that optimism using a stark warning from Louis Rossman.

00:10:42: Okay where does the optimism breakdown?

00:10:44: Rosamundt points out.

00:10:46: Yes, AI can write up PRD a product requirements document in two minutes.

00:10:51: But the PRD was never supposed to be the actual work

00:10:54: right?

00:10:55: The PRD is merely the receipt for the hard work of truly understanding a user problem.

00:10:59: Oh that Is A profound distinction

00:11:02: the receipt For the Work not the Work itself.

00:11:04: Exactly

00:11:05: The real work is wrangling stakeholders discovering the root cause of a workflow bottleneck and you know navigating the internal politics of deployment.

00:11:13: Yeah, AI can't do the politics for you

00:11:16: if he used claw to just generate the receipt without ever talking To the stakeholders or understanding why the current legacy process has broken.

00:11:23: You are totally disconnected from reality

00:11:25: Just generating paperwork?

00:11:27: He's have a very convincing document that provides cover for a terrible product decision.

00:11:32: A receipt proves you paid for the groceries.

00:11:35: Doesn't mean, You know how to cook a meal.

00:11:37: That is such good analogy

00:11:38: Which leads directly into critical question raised by Syed Tariq.

00:11:42: This raises an important Question.

00:11:45: By relying heavily on these tools To skip the painful discovery phase Are we accidentally breeding generation of AI tourists?

00:11:54: Tariq explains that a PM who only uses AI to write the requirements faster isn't transforming product management.

00:12:01: They are just accelerating administration.

00:12:04: So they're operating the toll booth faster without asking if cars even go into the right destination?

00:12:09: Precisely The truly valuable PM must do things the LLM cannot

00:12:14: Do.

00:12:14: Like what?

00:12:15: They have to validate demand in real world Build testable prototypes, navigate those human enterprise stakeholders you mentioned and tie decisions directly to revenue.

00:12:24: Right?

00:12:25: If you can't explain the AI technology but you cannot explain this specific customer value it creates Tarek says that your just an AI tourist.

00:12:33: So let's take these fast-built receipts And put them out in a wild.

00:12:39: What actually happens when these AI assisted products hit complex enterprise environments?

00:12:44: Oh!

00:12:44: That is where the hype just fractures.

00:12:48: Sean will periques here to cautionary tale about what he calls vibe coded sass alternatives.

00:12:53: Have you heard this term?

00:12:54: Vibecoded, yeah What is that?

00:12:56: it refers to software built by non-technical people who just keep prompting an AI until the interface looks right like It passes a vibe check on the front end.

00:13:04: Oh no But the back end is just a house of cards exactly.

00:13:08: Perique had a customer who migrated from an established sass platform To one of these ai build.

00:13:13: vibecoded tools looked super sleek

00:13:16: but did at work.

00:13:17: Soon enough, the tool started silently losing tasks and dropping comments.

00:13:22: The data integrity just totally evaporated

00:13:25: because an LLM can generate a beautiful react front-end in seconds.

00:13:29: but state management concurrency And relational database integrity those are mathematically rigid problems.

00:13:37: You can't just vibe your way to enterprise great architecture

00:13:40: you really can.

00:13:41: Perique even tested this himself by building a clone of his own software in Claude code.

00:13:46: It looked identical to the real thing!

00:13:49: But how did it perform?

00:13:50: Well, when he stress-tested it by dumping five hundred tasks into single project... The entire system just choked.

00:13:57: Simple actions started taking five seconds.

00:13:59: Scalability and data integrity are massive complexities that AI does not automatically solve for non technical owner

00:14:06: And lack architectural rigor pivots us directly into enterprise security.

00:14:10: Right, the security aspect.

00:14:12: Priyanka Upade argues that AI agent safety is fundamentally a product decision now not just an engineering or CISO problem.

00:14:20: How so?

00:14:21: Just because an AI agent has technical capability through an API to take action across connected enterprise systems doesn't mean it should be authorized to do so.

00:14:32: Well, it's like hiring an incredibly enthusiastic lightning-fast intern but handing them the master database password on their first day before teaching what they aren't allowed to touch.

00:14:41: That is exactly it!

00:14:43: If a user slips a malicious prompt into customer support ticket trying to trick AI in revealing another client's proprietary data The agent might technically have that database access

00:14:54: which is terrifying And that's why Apodye says you need deterministic controls designed into the system around of a model.

00:15:01: The AI shouldn't even have the option to bypass those boundaries,

00:15:04: right?

00:15:05: You need hard rules.

00:15:06: it requires a layered approach where product and engineering jointly decide were they agents.

00:15:11: probabilistic autonomy ends an hard-coded deterministic safety begins

00:15:16: taking all these enterprise complexities into account.

00:15:19: data architecture the deterministic security controls the strict boundaries.

00:15:24: Amit Shipuja makes a brilliant observation about where the real value sits today.

00:15:28: Okay, Where is The Value Shifting?

00:15:30: He says that competitive advantage is actually shifting away from AI models themselves and moving entirely to the context layer...

00:15:38: Because baseline models are basically becoming commoditized right?

00:15:42: An out-of-the box LLM might understand the general world but it doesn't understand your specific enterprise Exactly!

00:15:49: It does not know your historical architectural decisions your specific enterprise client base or you know, Your internal taxonomy.

00:15:56: Right?

00:15:57: So all those old PM artifacts we were just talking about the PRDs The roadmaps the decision logs that churned post mortems they aren't Just administrative paperwork anymore.

00:16:07: They are the proprietary training data for your context layer.

00:16:12: Oh wow so You're basically vectorizing your organizational history to train AI.

00:16:18: exactly The company with the most structured, accessible historical context layer is going to deploy the most effective safest AI agents regardless of whether they are using open AI and thropic or an Open Source model.

00:16:32: It completely revalues the documentation in history of a product organization.

00:16:37: You really does?

00:16:38: it's total paradigm shift.

00:16:40: And um...it actually brings us into final thought.

00:16:42: I want you the listener to mull over.

00:16:45: Let's

00:16:45: hear

00:16:45: it.

00:16:45: We've talked about speed generation the necessity of strategic filters, the evolution of the PM and the rigid reality of enterprise architecture.

00:16:54: Well Leary Hercules recently discussed the friction of taking things from a shiny AI prototype into actual production scale.

00:17:02: he noted that at AT&T's big data division it took a full thirteen months of support for their new products before the IT group was sufficiently trained to take them over an operate them.

00:17:13: Thirteen

00:17:14: months?

00:17:15: That is a staggering investment in time.

00:17:16: just hand over the keys.

00:17:18: It is, so here's a challenge for your own organization.

00:17:22: if you're team uses AI to build a genius working prototype in single afternoon does your enterprise actually have capability transfer plan place to operate govern secure and scale that product year from now?

00:17:36: That real test.

00:17:38: because speed of generation completely useless without an operational foundation.

00:17:43: I

00:17:45: mean, the x-ray machine might process this image faster than ever but you still need a specialist who knows how to read the murky results and prescribe actual treatment.

00:17:54: If you enjoyed today's episode new episodes drop every two weeks!

00:17:57: Also check out our other editions on cloud insights in Sovereignty, ICT & Tech Insights AI and Agentec Systems Sustainability In Green ICT DefenseTech And HealthTech.