Best of LinkedIn: Digital Products & Services CW 28/ 29

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

00:00:10: Frenness is a B to B market research company that supports enterprise product teams with jolting featurebyfeature competitive intelligence That shows exactly how their product stacks up against the competition.

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

00:00:23: So today's deep dive is specifically for digital transformation and tech professionals.

00:00:27: we are Basically bypassing all the usual hype cycle noise to look at top digital products and services trends currently circulating across LinkedIn.

00:00:37: We want to analyze what is actually surviving contact with real world right now in the trenches?

00:00:44: What if I told you that optimizing your team ship software faster, it's the quickest way to kill product today?

00:00:51: It

00:00:51: sounds so counterintuitive!

00:00:53: But really does.

00:00:54: but If we looked into stack of insights we've curated drop to zero and it is completely breaking how product organizations operate.

00:01:03: Yeah, the old rules just don't apply anymore

00:01:05: exactly.

00:01:06: So today we are going to unpack How AI is rewiring?

00:01:09: The Product Manager role from the inside out.

00:01:12: Why the discipline of deciding what to build Is suddenly way more important than how to build It and really where customer evidence fits into this whole new reality.

00:01:22: I love that the overarching mission here is really To understand the New survival metrics for product teams Because, I mean for the last decade value was tied directly to execution and velocity.

00:01:34: Right how fast can we get this out of door?

00:01:36: Exactly.

00:01:36: but when AI commoditizes execution The entire equation of what makes a human product team valuable it just flips upside down.

00:01:44: Well let's start with that structural rewiring then because for long time AI was treated as you know Just a feature like neat little chatbot.

00:01:51: You bolted onto an existing dashboard.

00:01:53: Oh!

00:01:53: For sure...just nice to

00:01:54: have.

00:01:55: But the sources were looking at indicate AI is no longer just the product, it's the production line itself.

00:02:01: Yeah to really ground this we can look at a twenty-twenty six Figma report that was highlighted by Rose B., The data essentially declares the death of traditional linear handoff….

00:02:10: …the

00:02:10: classic waterfallish passing of the baton.

00:02:12: Gone!

00:02:13: We are seeing rapid rise in what they call the multiplayer team and numbers found are honestly staggering like sixty percent of developers now actively participating in design.

00:02:22: Now sixty percent?

00:02:25: Forty-one percent of designers are actually writing code.

00:02:28: Wait, let's visualize this for a second.

00:02:30: the old assembly line was strictly sequential.

00:02:34: A PM writes a spec doc Tosses it over the wall to designer The designer makes a wireframe tosses into engineering and then they coat

00:02:42: Right.

00:02:42: Are you saying that entire factory floor has basically just dissolved?

00:02:46: Completely dissolved, I mean more than half of the teams in that survey have abandoned linear handoffs entirely because AI tools are bridging those technical skill gaps.

00:02:55: The work just moves in this continuous loop now

00:02:58: Like between code and design.

00:02:59: Yeah!

00:03:00: Between code canvas and critique.

00:03:02: a designer can literally prompt an interface directly into front-end code And a developer can adjust the visual architecture on the fly.

00:03:10: the mythical TenX individual development.

00:03:12: You know that whole trope?

00:03:13: Oh yeah,

00:03:13: The Lone Genius in a Hoodie

00:03:15: Right!

00:03:15: That's being replaced by the TenX team

00:03:17: Which really sets up an even more extreme scenario Outlined In A McKinsey Analysis shared by Angela Wick.

00:03:24: She points out that the entire software development life cycle, the SDLC is being rebuilt for AI to AI handoffs.

00:03:31: This where it gets really sci-fi?

00:03:32: It does because its not just designers and coders blurring lines anymore.

00:03:37: The AI agents are collaborating directly with each other.

00:03:40: Yeah...the mechanism she describes as fascinating.

00:03:42: So you have a product agent right And it ingests raw business objectives, user feedback.

00:03:48: All this telemetry data and it processes all of that to generate machine-readable users stories.

00:03:54: Then once a human PM approves the story It acts as a programmatic trigger That basically wakes up an architecture agent.

00:04:00: So one agent tags in next agent

00:04:02: Exactly!

00:04:03: The second agent consumes requirements And automatically generates API contracts, database schemas and system dependencies.

00:04:11: Let's

00:04:12: pause for anyone who isn't super deep into backend engineering.

00:04:15: An API contract is basically the menu of how two pieces of software are allowed to talk with each other.

00:04:21: Yeah, it defines inputs and outputs

00:04:23: And a database schema is just the blueprint for data stored.

00:04:27: I mean humans used to spend weeks arguing over those blueprints in meeting rooms Weeks?

00:04:32: Now that AI generates them instantly based on first agent's user story The human is no longer doing the manual translation between business need and technical architecture.

00:04:42: It really feels like we've stopped being camera operators on a movie set, and have suddenly been promoted to executive producers.

00:04:48: I like that analogy.

00:04:50: We aren't looking through lens or pulling focus manually anymore orchestrating the agents.

00:04:56: And that orchestration carries a massive financial premium right now.

00:05:00: if you look at compensation data from Akash Gupta, product managers who specialize in AI are now earning forty seven thousand dollars more than traditional PMS.

00:05:10: Forty-seven grand?

00:05:11: That is not a small bump!

00:05:13: Not at all.

00:05:13: but you know AIPM is NOT a buzzword title.

00:05:16: You can just slap on your resume to get a raise.

00:05:18: this pre-name is reserved for professionals Who have mastered six highly technical skills.

00:05:24: So I imagine we aren't just talking about knowing how to write a really clever prompt into chat GBT?

00:05:29: Oh, definitely not.

00:05:30: It requires knowing how to direct AI through complex multi-step work environments.

00:05:35: it's the ability to build a robust setup where the model actually has persistent context, memory and actual tools that can operate on its own.

00:05:43: That makes sense!

00:05:44: Furthermore...it is skill of writing programmatic evaluations before shipping.

00:05:49: You have to remember that A.I..is probabilistic

00:05:52: Right....It doesn't just do the exact same thing every single time.

00:05:54: Exactly.....its

00:05:55: not traditional calculator.

00:05:57: So you have measure & control the quality of those probabilistic outputs at scale rather than just kind of guessing if it worked.

00:06:06: The underlying models are basically a commodity now, and orchestration around them is an actual competitive

00:06:11: edge.".

00:06:20: Yeah, the goldfish memory problem.

00:06:22: Exactly!

00:06:23: So how are these highly paid PMs actually getting agents to hold on to contact?

00:06:27: Well

00:06:27: we have a really practical execution of this from Paoa Hurin.

00:06:31: He argues that you shouldn't treat an AI agent like a search engine.

00:06:35: you prompt You manage it like human direct report

00:06:39: Like an employee Right.

00:06:41: he uses specific markdown file A slaw ude dot md file To give his agent permanent Standing context that it reads before every single interaction.

00:06:51: Oh, so instead of typing out the company's background and current goals Every single time you open a chat The agent essentially reads an employee handbook Before he even answers you.

00:07:00: exactly

00:07:01: what kind of instructions is he putting in that file?

00:07:03: He hands the agent and okay are?

00:07:05: You know objectives and key results.

00:07:07: he tells It Exactly who the target audience Is and crucially What they're specifically not building this year.

00:07:12: That negative constraint is smart

00:07:14: Yeah, and he also defines its autonomy based on the concept of reversibility.

00:07:18: Meaning like if a decision is easily reversal say changing the color of a button on the dashboard?

00:07:24: The agent has the autonomy to just do it

00:07:26: right

00:07:27: but If It's irreversible Like dropping a core user database table the agent Is instructed To stop And demand human approval

00:07:33: Precisely.

00:07:34: He even embeds health metrics in the file that must not degrade while the agent chases its primary objectives.

00:07:41: The core takeaway from Huron is at.

00:07:43: the real skill of the future isn't prompting.

00:07:46: It's leading with deep structured context.

00:07:48: I can definitely see teams hearing this and immediately trying to buy like massive, super expensive all-in-one AI enterprise platforms to manage all these agents and workflows.

00:08:00: And that is exactly the trap Cheaton Saxena warns against.

00:08:03: Product leaders really shouldn't fall for the monolithic platforms that promise just do everything.

00:08:11: kind of situation.

00:08:12: Exactly,

00:08:12: Saxena actually ran a highly documented test.

00:08:15: he auditioned forty seven different AI tools over three months spent nearly twenty eight thousand dollars in the process and ultimately only kept nine of them.

00:08:23: wait you spend twenty-eight grand just to realize that thirty eight of those schools were essentially useless.

00:08:28: yeah what was the underlying flaw with?

00:08:29: The

00:08:31: ones that failed were usually bloated.

00:08:33: They tried to be a Swiss army knife and ended up being mediocre at everything.

00:08:37: the winning stack he landed on uses hyper focused specialized tools

00:08:42: like what?

00:08:43: So, He uses perplexity for deep market research granola For taking meeting notes without you know an obnoxious bot joining the call And Claude and Chad GPT for translating technical details.

00:08:55: Let me guess piece-mealing nine different tools together costs an absolute fortune in enterprise API fees.

00:09:01: Actually, no the total cost of his high performing stack is about sixty dollars a month.

00:09:06: oh wow yeah.

00:09:07: and the integration is incredibly rapid thanks to underlying technologies like Anthropics Model Context Protocol.

00:09:14: okay let's break down the model context protocol or MCP.

00:09:17: right from what I understand it acts as kind of like universal translator Right?

00:09:21: Because in the past, if you wanted an AI to read your local files or talk to a specific database.

00:09:27: A developer had to write custom plumbing for that exact connection.

00:09:30: MCP standardizes it right.

00:09:32: So the AI securely plugs into local data sources In hours instead of weeks.

00:09:37: That's perfect way to look at it.

00:09:39: But Sexena's ultimate conclusion is necessary reality check.

00:09:45: These tools dramatically reduce busy work, but they absolutely cannot replace human judgment.

00:09:51: Right like an AI generator.

00:09:53: priority list is a great starting point But the humans still has to make the final call

00:09:58: which really transitions us into our next point.

00:10:00: Yeah because if AI makes the actual building of software incredibly fast and essentially cheap It fundamentally shifts the bottleneck for every product team out there.

00:10:10: yeah

00:10:11: it moves upstream

00:10:11: exactly.

00:10:12: The challenge is no longer how to build.

00:10:14: Mm-hmm, the challenges deciding what to built

00:10:16: on a percent

00:10:17: and hey real quick.

00:10:18: by the way if you want To keep catching these enterprise tech trends before they hit the mainstream Hit subscribe so you don't miss our future deep dives.

00:10:25: Yeah definitely subscribe.

00:10:26: but getting back to that shift from Howe to What this brings us?

00:10:30: Insights form Shardal Mepha and Basia Kubica.

00:10:33: They observed that for the last fifteen years product managers have been completely obsessed with agile methodology scrum ceremonies and just maximizing sprint velocity.

00:10:43: Just going as fast as possible?

00:10:44: Yeah, right!

00:10:45: The entire industry was geared toward optimizing the software assembly line but... Today, because basically anyone can build a feature in an afternoon the result is massive graveyard of unused features.

00:10:58: The Feature Factory's working at light speed but it just pumping out code nobody actually wants to use.

00:11:03: Exactly Shipping the wrong thing faster isn't progress It's accelerated waste.

00:11:08: Accelerated

00:11:09: Waste is great way put it.

00:11:10: Therefore the most scarce and valuable skill In product organization no longer execution Yeah.

00:11:18: AI can generate infinite variations of a user interface, but knowing when to

00:11:22: stop

00:11:23: or knowing what not to build that requires a human understanding of unit economics and actual business outcomes.

00:11:29: Jacqueline Consulman has great term for this.

00:11:31: she calls it product taste in restraint.

00:11:34: just because the new customer requirement shows up on your inbox Or Just Because You Physically Can Build A New Feature By Tomorrow Using An Agent doesn't mean you should bolted onto Your Product.

00:11:43: It's So Tempting Though!

00:11:45: But Because Building Isn'T Hard Anymore Shipping is no longer the ultimate signal of competence.

00:11:51: The discipline to hold back and rethink this shape what you are making, that's a new marker for real product taste.

00:11:59: I completely agree with it

00:12:00: But i have to play devil's advocate here For A Second.

00:12:02: That sounds beautiful in textbook but how do you defend that restraint In a boardroom?

00:12:10: When the C-suite is looking at competitors launching AI features every single week and they are demanding that you ship faster to keep up.

00:12:20: How does a product manager practically justify saying no without getting fired?

00:12:25: It is arguably the single hardest part of the job right now, and Alina Leanova provides a framework for surviving that exact meeting.

00:12:32: Okay let's hear... She argues that at the executive level profit strategy is not a roadmap of features.

00:12:38: it is an exercise in capital allocation.

00:12:40: So when you bring a proposal to the C suite saying, you know customers asked for it or competitors have It that is a completely insufficient argument

00:12:48: because you're asking for money

00:12:49: Exactly.

00:12:50: You are asking for scarce company capital and engineering headcount.

00:12:54: so what does the actual?

00:12:55: Mechanism to enforce restraint in that meeting?

00:12:58: you have

00:12:59: to force The uncomfortable question you literally have to ask What gets worse because we made this choice?

00:13:05: well That's good

00:13:06: right.

00:13:06: if you cannot answer that If you claim that nothing is sacrificed and we can just do it all magically, then you don't actually have a strategy.

00:13:15: You just have a wish list!

00:13:16: True strategy requires explicit painful trade-off.

00:13:19: Yes...you

00:13:21: have to explain the C-suite what alternative you actively rejected And What specific risk are accepting by choosing this path over another?

00:13:29: Okay but I could hear a CEO responding to that saying Nothing gets worse…I want the new feature, fast, cheap and perfect figured out.

00:13:36: Typical CEO response.

00:13:38: Right, so how does a PM handle the stakeholder who just flat out refuses to accept trade-offs?

00:13:45: That's where you lean on Ali Baruti strategy.

00:13:47: regarding scope We are all familiar with the classic project management triangle right cost quality and speed.

00:13:54: The old joke is that you only get to pick two

00:13:56: exactly but Baruti argues that costs quality and speed don't actually have to compete against each other as long you are ruthlessly disciplined about your scope.

00:14:06: Basically applying the Pareto principle, roughly eighty percent of the actual customer value is going come from twenty-percent of features that you build

00:14:13: Exactly!

00:14:14: If you had the discipline stop trying building the remaining eighty percent which lets face it usually edge cases.

00:14:21: just add technical debt complexity and defects.

00:14:24: You can deliver high quality software very quickly well under budget.

00:14:29: So the trade-off isn't speed versus quality?

00:14:32: No.

00:14:32: The trade off is refusing to build the bloated scope that the CEO initially asked

00:14:36: for, and that restraint applies directly to how we view AI itself too.

00:14:41: Mauricio Cardenas points out that AI is a capability choice not a product strategy in of itself.

00:14:47: That's so important to remember.

00:14:48: You have to match this specific underlying technology To the specific customer problem you were trying to solve.

00:14:56: This is a critical distinction, because if the customer problem is say forecasting inventory based on three years of historical data.

00:15:05: You don't need a massive expensive large language model for that.

00:15:09: you need traditional deterministic machine learning.

00:15:12: Let's clarify that difference really quick.

00:15:14: Deterministic software was like a calculator.

00:15:16: If you put in two plus to it will always equal four.

00:15:19: Traditional Machine Learning can look at past numbers and cleanly predict future Numbers.

00:15:24: right

00:15:25: but an LLM is probabilistic.

00:15:28: If you ask it two plus, too It might say four and might say for spelled out.

00:15:32: or Mike just apologize for not being a math tutor You wouldn't use a probabilistic tool to build a strict deterministic compliance rule

00:15:39: which Is exactly why bolting an LLM on everything as a terrible strategy.

00:15:43: now if the user workflow involves summarizing messy unstructured documents then yes An LLm is the perfect capability choice

00:15:50: because it excels at ambiguity.

00:15:52: Yes

00:15:53: Customers are not buying the label AI.

00:15:55: They're buying faster resolution times and lower risk.

00:15:59: The strongest product teams won't be the ones with the most ai features.

00:16:03: they will Be, the ones that apply?

00:16:04: The right technological capability to the right workflow.

00:16:07: so if we pull this all together If supreme judgment intense restraint And smart capital allocation Are the new core skills of a product manager where does that judgement actually come from?

00:16:18: That's the million dollar question

00:16:20: because you can't just walk into a boardroom and enforce major scope trade-offs based on your gut feeling.

00:16:25: No, it has to be grounded in unshakable customer evidence.

00:16:29: if your restraint isn't backed by data You're basically just guessing faster than the competition.

00:16:34: but The

00:16:34: problem right now is that teams are kind of drowning In data.

00:16:37: Teresa Torres has a major warning about this exact dynamic.

00:16:41: What does she see happening?

00:16:42: Well teams have access to behavioral analytics dashboards thousands of support tickets app store reviews transcripts from sales calls.

00:16:51: It is an absolute mountain of information,

00:16:53: right?

00:16:53: More data than ever.

00:16:54: but instead of finding the truth they are projecting their own internal bias onto these low quality signals.

00:17:00: They think you're doing objective customer research But there really just cherry-picking data points to justify building The feature that I already wanted to build in the first place.

00:17:09: Oh confirmation bias at scale.

00:17:11: she calls this the ladder Of evidence.

00:17:13: Right!

00:17:14: The fundamental rule Is That as the effort required To get the data goes up.

00:17:17: The strategic value of that data also goes up.

00:17:20: Meaning a casual stakeholder meeting or product demo where you just ask the client, hey do like this interface?

00:17:26: Not valid customer research.

00:17:29: Torres argues for prioritizing rich story-based interviews over mere volume.

00:17:34: So quality over quantity.

00:17:35: Yes A mediocre one on one interview Where actually listen to real friction in customers?

00:17:41: daily workflow is vastly superior to having an AI.

00:17:45: just summarize a thousand low-contact support tickets.

00:17:49: We also have to look at who we are getting this evidence from.

00:17:51: Rishabh Gupta brings forward a deeply counterintuitive angle on those.

00:17:55: Okay, Who Are We Missing?

00:17:56: Well in product management teams absolutely obsess over the new user activation funnel.

00:18:02: They spend billions of dollars running AB tests trying find that magical aha moment In The First Seven Days Of A User's Journey.

00:18:09: But in doing so they completely ignore the exit.

00:18:12: You mean the churn like the people who actually cancel.

00:18:15: Yes, somewhere in your user base a loyal customer used you product every single day for two years just quietly churned last month.

00:18:23: Wow!

00:18:24: They didn't write an angry email they did not file support ticket.

00:18:29: That specific user knows exactly where your product stopped keeping its promises.

00:18:34: They know the exact moment that friction in software stops being worth a price, or they knew which competitor finally built something good enough to make them switch Precisely.

00:18:42: We spend all our budget analyzing users who never even converted and almost nothing trying to understand psychology of loyal users walking away.

00:18:52: Studying the exit is when real-product truth lives.

00:18:55: So if we gather this truth, how do we actually test it in an AI-driven world?

00:19:01: VJJ presents a completely new approach to prototyping.

00:19:04: Because the old way doesn't work anymore

00:19:06: Right!

00:19:06: For years PMs built wireframes and clickable mockups to validate visual interface.

00:19:12: You wanted know did navigation make sense?

00:19:15: was layout intuitive?

00:19:17: That worked perfectly for traditional deterministic software where clicking button A always leads screen B.

00:19:24: But as just discussed AI is probabilistic.

00:19:27: The uncertainty isn't an individual interface anymore.

00:19:30: The uncertainty lives in the ambiguity of human

00:19:32: behavior.".

00:19:33: Exactly!

00:19:33: If a AI agent is going to execute a multi-step workflow for a user, a clickable wireframe is basically useless – it can't simulate agents' logic or weird edge cases….

00:19:43: So what's VJ's solution?

00:19:45: He argues that in the A.I era prototypes shouldn't exist to validate an interface.

00:19:50: They exist to invalidate your assumptions about human behaviour when intelligence is injected into real messy workflow.

00:19:57: He calls the production prototype, right?

00:19:59: Instead of building a fake interface you put raw AI behaviors directly into the tools your team or users are already communicating in.

00:20:07: You watch what actually changes and their behavior.

00:20:09: does The AI actually reduce their context switching Or is it quietly introduce some brand new type of friction?

00:20:16: Right.

00:20:16: You have to invalidate your assumptions about how humans interact with probabilistic logic before you authorize writing expensive production code, which

00:20:24: brings us the most raw honest behavioral data.

00:20:27: a company actually has customer support.

00:20:29: Oh for sure.

00:20:30: Jerry Kobilka and Ahana Banerjee both highlight that support conversations are an absolute goldmine of evidence.

00:20:38: Cobelco notes an incredible metric.

00:20:40: Ten conversations on a platform like Intercom will yield an average of eight actionable product insights.

00:20:45: That ratio is phenomenal, eight out of ten It's

00:20:47: massive.

00:20:48: But historically it has been incredibly dickapult for product managers to extract that clear signal from the overwhelming noise Of angry confused and highly emotional users.

00:20:58: And this is where AI fundamentally changes the game for discovery.

00:21:02: Support isn't just a cost center meant to close tickets anymore, it's real-time user research.

00:21:08: but you know human users are vague.

00:21:10: they're frustrated and often ask like five unrelated questions in single chat message.

00:21:16: Yeah

00:21:16: good luck parsing that manually.

00:21:18: But AI can finally scale the listening process required to make sense of that chaos.

00:21:23: You can use models to ingest thousands of messy emotional support chats and structure them into quantified product feedback

00:21:30: without losing a nuance.

00:21:31: Exactly!

00:21:32: The AI can identify exactly where the product is causing confusion, categorize the severity ,and then route highly sensitive issues to humans.

00:21:41: all with out loosing empathy are actually require help customer in moment.

00:21:44: It essentially allows the human support agents to focus on complex emotional resolution while feeding pure structured behavioral signal directly back to the product team so they actually know what to build next.

00:21:55: Yeah, and if we step back and look at the trajectory of everything we've analyzed today it leads to a pretty profound realization.

00:22:02: What's that?

00:22:02: Well... If AI agents are rapidly taking over the how of coding testing in building And The Human Product Manager's role shifts entirely into the what-and-the why The PM of twenty-twenty six is essentially acting as a behavioral economist and a capital allocator wrapped into one.

00:22:19: That's

00:22:19: a huge shift in identity, it

00:22:21: really is.

00:22:22: but here something to mull over before we wrap up.

00:22:25: if AI can perfectly read a messy customer support ticket diagnose the friction And then automatically generate the exact code required to fix It how long until the human product manager isn't just reshaved But bypassed entirely by system that links the users complaint directly to the Codebase?

00:22:42: Ask yourself, are you managing the decisions or just a temporary

00:22:56: bridge?