Best of LinkedIn: Digital Products & Services CW 44/ 45

00:00:00: This episode is provided by Thomas Allgaier and Frentice, based on the most relevant LinkedIn posts about digital products and services from CW-IV & IV.

00:00:11: Frentices 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.

00:00:23: You can find more info in the description

00:00:25: Right!

00:00:25: And um... To kick things off today we are mapping out the absolute top digital products and services trends that are currently dominating tech circles on LinkedIn.

00:00:34: Yep,

00:00:34: exactly.

00:00:35: And the mission for this deep dive is to really explore how The rapid acceleration of AIs just fundamentally rewiring How software gets built.

00:00:44: I mean it's changing what product managers actually do all day.

00:00:47: Oh absolutely!

00:00:48: We'll get into how To measure real AI ROI and how companies Are basically being forced to restructure their operating models entirely Just to survive

00:00:57: Because a foundational shift here The thing driving all of this is the sheer velocity of delivery.

00:01:04: Execution speed is essentially becoming commoditized,

00:01:07: right?

00:01:07: Which Arnaud Beaux pointed out in a recent post.

00:01:10: he highlighted a macro shift that really puts us into perspective.

00:01:14: if you look at the timeline of software delivery the cycles have just completely collapsed.

00:01:18: Yeah like think back to the boxed software era You know when you had to physically press a program onto a CD-ROM

00:01:26: Yes, and then ship it to a literal store.

00:01:28: I mean shipping took years And you couldn't afford a single bug.

00:01:32: because well patching It was nearly impossible

00:01:35: right?

00:01:35: You had one shot to get it Right But didn the sass air arrived and shrink that feedback loop down too?

00:01:42: what those standard two-week agile sprints?

00:01:44: Exactly.

00:01:45: Two weeks was the gold standard, but today because of continuous telemetry and generative AI models just writing code on the fly our knob notes those iteration cycles have collapsed down to mere hours

00:01:58: which is just wild to think about.

00:02:00: But it forces us to look at these structural mechanics of a tech company.

00:02:04: if execution speed is near instant The bottleneck in the whole system has to move.

00:02:09: Yep.

00:02:09: And Sean G and Nayan Teja both highlighted this exact shift.

00:02:14: because AI coding agents are accelerating the actual keystrokes of writing code so drastically, that constraint is no longer the engineering output.

00:02:23: So it's not about how fast you can type the code anymore?

00:02:25: Exactly The new bottlenecks our requirements verification in human judgment...the whole organizational system starts choking on decisions Not on code generation.

00:02:38: Okay Let's unpack this a bit because.

00:02:41: This leads to really provocative question if the code is that easy to generate and a founder can just, you know explain it problem-to-clawed.

00:02:49: get a custom app built over weekend.

00:02:52: doesn't that completely destroy traditional twenty nine dollar month SAS subscription model?

00:02:58: Alex Talander actually argued this exact point.

00:03:01: why pay a vendor for rigid platform when an AI can build you bespoke tool in five minutes?

00:03:07: I mean, it looks like a total existential threat to SAS on the surface.

00:03:10: But if we break down the operational physics of that scenario Tom Colopy points out a massive hidden contradiction.

00:03:18: How so?

00:03:19: Well let's follow the logic.

00:03:21: say uh SAS vendor has a thousand enterprise customers and those a thousand customers look at GenAI And decide hey We are going to replace this vendor with our own custom-built AI

00:03:33: tools.

00:03:34: Okay, so everybody gets their own weekend built hyper-customized application?

00:03:38: Sounds great for them.

00:03:40: Right it sounds great but software doesn't just sit in a vacuum once its deployed.

00:03:45: the environment around is constantly changing.

00:03:47: dependencies update third party APIs get deprecated security vulnerabilities pop up Oh

00:03:53: and AI prompts themselves to suffer from drift over time.

00:03:55: right Exactly Suddenly each of those a thousand customers realizes they need to hire one or two dedicated developers just to maintain their bespoke AI app and direct the agents to fix breaking changes.

00:04:07: Wow, okay.

00:04:08: so you've essentially created thousands of new developer jobs in massive operational overhead across all those individual companies?

00:04:16: Exactly!

00:04:17: The opposite of saving money...the aggregate cost of maintaining THOUSANDS of custom apps vastly exceeds original vendor's centralized development costs.

00:04:26: It just reinforces this brutal law of technology.

00:04:29: Writing the initial code is always the easiest part.

00:04:32: Maintaining the software and securing it over a decade, Is The Incredibly Difficult Part!

00:04:36: It's like putting up the framing of house for free but then realizing you have to do all your own plumbing & electrical maintenance by yourself forever.

00:04:42: Perfect analogy...the burden shifts.

00:04:45: So if the bottleneck shifting heavily toward decision making and maintenance How does actually altering day-to-day workflow of product manager?

00:04:54: Because Clairvaux shared this wild example about what's happening over at Figma.

00:04:58: Oh yeah, the Figma setup is fascinating!

00:05:00: It

00:05:00: really is... They are running a thirty-one skill PMOS built entirely on cloud code and The crazy part that it all runs from one Shared Git repository just to keep sixty to seventy product managers totally in sync

00:05:14: which honestly tells you just how technical the product management discipline is becoming right now.

00:05:19: because Git as a version control system historically used strictly by software developers, to track changes in code.

00:05:25: Right it's a dev tool

00:05:26: Exactly.

00:05:27: but because PMs are essentially writing logic through these complex AI prompts they have to use developer tools to version-control their product strategy.

00:05:35: They're treating requirements like living code base.

00:05:39: But doesn't that blur the line entirely?

00:05:41: If PMs are using developer environments to manage AI agents, they're basically acting as systems engineers at this point.

00:05:48: They totally are!

00:05:50: And in some environments the human engineering layer is being bypassed entirely.

00:05:54: Ben-Sara shared his setup at Policia which a fully autonomous pipeline.

00:05:58: Wait really?

00:05:59: Fully Autonomous?

00:06:01: Yeah.

00:06:02: During live customer chats, an AI agent actively logs issues based on the conversation.

00:06:07: Then a completely distinct PM agent reads and ranks the day's tickets based on priority parameters.

00:06:13: Okay so AI is doing tree dodging

00:06:15: Yep.

00:06:16: And then top items bypass human backlogs completely.

00:06:20: They go straight to engineering agents.

00:06:21: that actually writes code for the fix.

00:06:25: An AI coding model like Codex steps into QA The logic before humans ever really looks at it.

00:06:31: But having an AI QA its own generated code feels incredibly risky.

00:06:36: Like how do you trust that verification loop if a human isn't testing the boundaries of the fix?

00:06:41: Well, that is the vulnerability in the loop.

00:06:44: right The assumption is that Codex being specifically trained for code evaluation can spot the logic flaws but it still requires a human at the end of the chain to approve the merge

00:06:54: Still...the speed undeniable.

00:06:57: Oh, absolutely!

00:06:58: They are identifying a bug in a chat and shipping the fix almost in real time which

00:07:02: actually brings up The very beginning of that entire loop product discovery Like how do we effectively listen to the user if AI is doing all the building and triage?

00:07:11: Right That's the big question.

00:07:13: Yeah And Kapil Devarumushan pointed out that AI models can now effortlessly extract repeated frustrations buying signals and feature requests from entirely scattered customer touch points.

00:07:23: Which solves a massive historical problem for teams?

00:07:26: It does, because usually sales hears one thing support hears another and product hears the third.

00:07:32: but I mean how does the AI handle conflicting feedback if sale says that features too expensive But supports as it's to complex?

00:07:40: How is model resolved without hallucinating some weird middle ground?

00:07:44: Well honestly often doesn't resolve well which is exactly why Beth McHugh pushed back heavily on this during her recent panel at Product Fest, twenty-twenty six.

00:07:53: Oh what was her take?

00:07:54: Her argument is that AI can actually degrade the quality of product discovery because it creates this false sense of deep understanding.

00:08:01: oh I see... Because if a team just reads in neat little bulleted summary and assumes they know exactly with customer wants

00:08:07: Exactly And humans naturally anchored to first plausible idea So teams use AI to skip the messy part of discovery and go straight into ideation based on a summary.

00:08:18: They end up iterating toward a slightly less bad version of a fundamentally flawed premise.

00:08:22: That makes total sense.

00:08:24: And McHugh also pointed out that AI fundamentally lacks an understanding of human nuance.

00:08:28: like it cannot tell when a customer is giving a quote.

00:08:31: polite lie during an interview

00:08:33: A polite lie

00:08:34: Yeah, you know what they say.

00:08:36: Oh, that's interesting.

00:08:37: Just to be nice versus actually showing real intent to buy.

00:08:40: she even cited instances where the AI literally fabricated fake quotes just to make it summary sound more cohesive.

00:08:47: Wow polite lies are this silent killer of good product development?

00:08:52: building a roadmap on hallucinated enthusiasm is just a recipe for disaster.

00:08:56: by The way if you our product manager listening right now and feeling the ground sort of shift under your feet.

00:09:02: Make sure to subscribe to this show so you don't miss our future deep dives!

00:09:06: The mechanics for that job are changing weekly.

00:09:09: Yeah, definitely subscribe because if AI is actively dangerous when applied to Deep Human Discovery Laura Shores offered a much better framework where PMs should actually deploy it.

00:09:20: Okay, where does she think it belongs?

00:09:21: She says AI should not be tasked with empathizing.

00:09:25: It should be tasked With absorbing the crushing administrative burden.

00:09:28: You know...the documentation The meeting summaries Cross-referencing JIRA tickets.

00:09:32: Oh absolutely Taking away the busy work

00:09:34: Right.

00:09:35: If the AI handles the bureaucratic way Product managers can actually return to the core of the discipline Human empathy Unstructured curiosity And exercising strategic judgment In the room with a customer.

00:09:47: Reclaiming time for empathy sounds incredibly noble, but let's look at the actual business impact here.

00:09:53: We are accelerating delivery we are automating admin But are we actually making any money?

00:09:59: That is a million dollar question right?

00:10:02: Grace gong highlighted and massive ROI reckoning happening right now.

00:10:07: Roughly ninety five percent of generative AI pilots Are currently falling below measurable business impact.

00:10:12: ninety-five percent

00:10:14: yeah

00:10:14: that huge.

00:10:16: And CFOs are staring at these massive enterprise AI compute bills

00:10:20: and

00:10:21: They're only giving product teams a runway of about twelve to eighteen months To prove the experimentation was actually worth it.

00:10:27: Well, The problem there is the metrics teams Are using to justify that massive compute spend.

00:10:32: Vin Vashista dissected this perfectly.

00:10:34: he sees product team celebrating things like A fifty six percent increase in tasks completed by their users which

00:10:41: sounds great on paper

00:10:42: It does.

00:10:43: But task completion is a productivity metric.

00:10:45: It's not an ROI metric.

00:10:48: Productivity only matters if it translates to revenue growth or higher free cash flow,

00:10:52: right?

00:10:53: If CFO sees a fifty-six percent efficiency gain without a corresponding bump in revenue they are simply going use that efficiency metric as the justification to cut headcount and maintain margins.

00:11:05: Oh wow!

00:11:06: Okay let me test an analogy here to see if I'm understanding.

00:11:12: Tim Herbig used this concept of Used Car Marketplace.

00:11:14: Okay,

00:11:15: I like that one!

00:11:15: Yeah it seems most companies are just slapping a generic chat with our AI button onto their interface and expecting a revenue bump.

00:11:23: but if you put a generic assistant on a car marketplace its functionally useless

00:11:27: Completely useless.

00:11:28: Right because the first time car buyer needs the AI to calculate monthly affordability insurance rates basic running costs.

00:11:35: But a veteran car buyer needs that exact same AI to instantly cross-reference maintenance histories and appreciation curves across specific trims.

00:11:43: Exactly!

00:11:44: If the AI doesn't dynamically adjust who it is talking, It provides zero commercial value.

00:11:50: And THAT IS THE CORE ISSUE RIGHT THERE.

00:11:53: CONTEXT IS THE ACTUAL PRODUCT NOT THE AI AGENT ITSELF.

00:11:57: Jennifer Sherman hammered this home regarding enterprise resource planning software or ERPs, you know the massive database systems that act as a financial nervous system for large companies.

00:12:06: Right they're really complex enterprise stuff.

00:12:09: Yeah She explained that true AI value comes entirely from deeply understanding The specific industry context of the customer.

00:12:16: How's it play out in the data though?

00:12:18: We'll take a standard row and an ERP database labeled project.

00:12:22: if your customer is a corporate consulting firm project row represents a client engagement with the very strict profit margin target.

00:12:29: So, the AI needs to optimize for billable hours...

00:12:32: Makes sense!

00:12:32: But if your customer is an on-profit organization that exact same Project Row is actually restricted.

00:12:38: grant A government funder watching every single dollar ensure strict regulatory compliance.

00:12:43: Profit isn't goal zero variance compliances

00:12:46: Oh, I see.

00:12:47: Right.

00:12:48: so if a generic AI agent treats those two database rows the exact same way it is going to execute massive unrecoverable financial mistakes faster than a human ever could?

00:12:58: So that's not about how fast an agent can process data.

00:13:01: It's about the agent deeply understanding of the unwritten rules or specific game its playing.

00:13:05: Precisely

00:13:06: Dr Bartjaorski actually brought up consumer example that illustrates what looks like when you get this right.

00:13:12: He analyzed MAL which was an AI native bank.

00:13:15: Their agent doesn't act like a passive chatbot that just visualizes your account balance.

00:13:20: Okay, what does it do?

00:13:21: You set a strict financial budget up front and the AI agent will actively leave the banking environment go out to an airline And book a flight for you operating strictly within The Financial Boundary you authorized.

00:13:33: It solves the user's job end-to-end.

00:13:35: That's incredible.

00:13:37: But you know delivering that kind of context aware End-To-End execution requires completely different organizational structure.

00:13:44: You cannot build highly integrated, boundary pushing AI products if your company is trapped in legacy organizational silos.

00:13:51: Which means we have to talk about product operating models.

00:13:54: you just can't ship contextual AI If your finance and engineering teams are still budgeting like it's twenty fifteen.

00:14:00: Definitely not.

00:14:01: Melissa Perry pointed out a structural flaw In the industry here.

00:14:04: Most companies Are running an undesigned Operating model.

00:14:08: No one purposefully architected how the business functions.

00:14:11: The annual funding process was dictated by finance a decade ago, engineering structure is a hangover from a reorg three years ago

00:14:18: and the actual product strategy's dictated By whatever executive is shouting the loudest in the room

00:14:23: exactly to structure just awkwardly accumulated over time.

00:14:27: And when leadership realizes they are moving too slowly They usually default To what Jen Swanson calls transformation theater transformation.

00:14:35: I love that term.

00:14:36: it's

00:14:37: so accurate.

00:14:37: They buy new agile software tools, they introduce OKRs and they just rename all their project managers to product managers.

00:14:45: But they don't actually alter the underlying physics of the organization.

00:14:49: It's like bolting a slick Tesla autopilot dashboard onto one nineteen nineties diesel tractor.

00:14:55: The interface looks incredibly fast in modern but underneath the engine still takes ten minutes to change gears.

00:15:01: That is the perfect way to put it.

00:15:03: Funding still happens on rigid annual cycles, technical decisions have to crawl through five layers of middle management approval.

00:15:10: The terminology changes but the actual flow of work remains completely static.

00:15:15: So are there any documented examples of a massive enterprise actually fixing this structural friction?

00:15:21: There are...but they're rare.

00:15:22: Ava Lahon shared a success story from AXA's one GTO tech factory.

00:15:27: They fundamentally industrialized their operating model by breaking down the silos between products, service operations and local markets...

00:15:34: And they didn't just change titles

00:15:36: right?

00:15:36: No!

00:15:36: They physically stripped the funding and decision-making power out of legacy middle management layers and tied it directly to three strict priorities.

00:15:45: They refocused solely on capabilities that create direct value industrialized delivery through reusable platforms, and aggressively flatten the hierarchy to accelerate decisions.

00:15:56: Yeah they integrated the actual financial accountability right into the product teams.

00:16:00: Accountability is the crucial variable there especially when you look at how these empowered teams are actually staffed.

00:16:06: Saba Quasi & Kunal Doudraj Biswakarma highlighted a fascinating data point from levels.fyi about the ratio of PMs to engineers across big tech.

00:16:16: Oh, the ratios are wild.

00:16:17: They're wildly inconsistent.

00:16:19: at Apple The ratio is roughly one PM for every thirty engineers.

00:16:23: but a Walmart it's one p.m.. For every three point six engineers.

00:16:27: So what does this all mean?

00:16:29: who actually has the correct operational ratio?

00:16:31: Well the ratio itself Is a meaningless metric unless you understand the context of the product being built.

00:16:36: okay have some.

00:16:37: if a team is maintaining A highly mature internal platform with a clearly defined technical backlog, a single PM can easily manage the prioritization for thirty engineers.

00:16:46: Right

00:16:46: because the path is clear?

00:16:47: Exactly but if a team has tasked with exploring a brand new consumer journey that requires massive qualitative discovery constant prototyping and intense cross-functional alignment with legal in marketing then that PM will be completely maxed out.

00:17:02: supporting just three or four engineers.

00:17:04: That makes total sense.

00:17:06: Anansagar broke this down by defining six completely different types of product managers.

00:17:11: He argued that trying to hire a generic PM is like trying to buy a generic vehicle for a construction site.

00:17:16: Yes, I saw that.

00:17:17: yeah

00:17:18: Is it?

00:17:18: if you hire a growth Pm?

00:17:20: You are hiring a formula one car.

00:17:22: They are obsessed with acquisition loops conversion speed measurable revenue.

00:17:27: But a platform.

00:17:28: pms was an excavator.

00:17:30: they're methodically building the deep robust infrastructure That every other team relies on.

00:17:35: right You cannot judge an excavator by its top speed on a racetrack, and you can't evaluate platform PM using growth metrics.

00:17:42: Yet companies make that specific hiring mistake constantly—you know they hire a highly aggressive GrowthPM to solve a deeply technical platform infrastructure problem…and then wonder why the architecture collapses one year later?

00:17:55: Exactly!

00:17:56: Furthermore as AI completely removes friction from writing code...the sheer volume of features being shipped is skyrocketing.

00:18:04: Jesleen Kaur noted that this velocity is driving the rapid rise of product operations.

00:18:09: Because someone has to manage all that output?

00:18:11: Right!

00:18:11: Product ops have to step in as the central radar and conductor, because if AI allows an engineering team ship ten features a week but the sales team doesn't know what's launching.

00:18:22: leadership does not know what business outcome is supposed be.

00:18:26: fast execution just devolves into fast chaos.

00:18:29: Fast Chaos I mean, the code is flying out of door.

00:18:33: The dashboards are green but overarching business strategy completely disjoint it.

00:18:38: Exactly!

00:18:39: Which brings us to ultimate takeaway for people actually doing their work in these trenches We've covered collapsing delivery cycles, harsh reality AI ROI and desperate need to redesign how teams operate.

00:18:52: When you synthesize all this source material what's core lesson for individual contributor?

00:18:58: Well, Ron Yang and Roshan Gupta summarized the shift perfectly.

00:19:03: The real value of a product manager in frankly any technology leader was actually never about overseeing the build process Right

00:19:10: managing to sprint board and over seeing the code delivery.

00:19:13: it's just a bureaucratic task.

00:19:15: It is essentially a massive administrative distraction from the actual job.

00:19:20: So what is the actual Job now?

00:19:22: It has always been influence without authority.

00:19:25: If Generative AI eventually drives the marginal cost of writing and deploying code down to practically zero, The only mechanism that will keep a company alive is human judgment.

00:19:34: Just knowing what to do?

00:19:35: Yes!

00:19:36: It's the ability.

00:19:37: look at a chaotic market, decide exactly WHAT IS WORTH BUILDING, align a highly fractured organization behind specific vision... And empower go-to-market teams actually sell it.

00:19:49: When technical execution becomes a cheap commodity, strategic judgment is the only defensive moat a company has left.

00:19:55: Judgment Is The Moat.

00:19:57: It's no longer about how fast you can type the prompt.

00:20:00: it's About having deep contextual awareness to know which prompt actually worth typing in first place.

00:20:06: If enjoyed this episode new episodes drop every two weeks.

00:20:09: Also check out our other editions on cloud insights and sovereignty ICT & Tech Insights AI in Agentec Systems, Sustainability and Green ICT, Defense Tech & HealthTech.