Best of LinkedIn: Digital Products & Services CW 18/ 19

00:00:00: This episode is provided by Thomas Allgaier and Frennus, based on the most relevant LinkedIn posts about digital products & services in calendar weeks eighteen-nineteen.

00:00:09: 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.

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

00:00:23: Yeah, and today we are taking a really deep dive into exactly those trends.

00:00:28: We've got a ton of great insights from across the ICT in tech industry over the past couple weeks

00:00:33: Right!

00:00:33: And mission for this Deep Dive is well it's honestly about synthesizing something profound that happening right now... ...we're looking at how AI completely rewiring role as product

00:00:43: manager Exactly It shifting everything From How we build to what actually built all way down to very users were building massive shift.

00:00:53: It really is, so let's start with that first thing which is the sheer velocity of AI native product building.

00:00:59: there's this huge debate happening right now on LinkedIn about The Death Of The PowerPoint PM.

00:01:04: Oh yeah!

00:01:04: The classic Powerpoint PM.

00:01:06: Right because traditionally a Building A Digital Product felt a bit like running This highly complex slightly bureaucratic assembly line.

00:01:14: you had the planners sketching the blueprints then designers making them mock-ups and finally engineers writing the code?

00:01:20: Yeah it's sequential.

00:01:21: Here it relies entirely on handoffs, and every time an idea moves from say a product manager to designer you lose time.

00:01:29: You just add this whole translation layer

00:01:32: Exactly.

00:01:33: But he noted that AI basically removes the blank page problem entirely.

00:01:41: The blank page problems, right?

00:01:42: Because initial momentum is the hardest part.

00:01:45: Yeah and he described going from a raw idea straight to product requirements document down into prioritized backlog And then finally working prototype in

00:01:56: single day.

00:01:56: I mean, think about the mechanics of what Hisama is describing there.

00:01:59: In the past writing a PRD required you to sit down stare at blank screen and manually type out every single user story.

00:02:05: in edge case

00:02:06: it was exhausting

00:02:07: totally exhausting.

00:02:09: now The AI generates that initial mass of text so the product manager shifts from being this Exhausted author two being an editor.

00:02:18: And we're seeing this play out in the real world too.

00:02:20: Srinivasan Raghavan shared an example from his work at Freshworks that honestly blew my mind.

00:02:25: Oh,

00:02:26: the Freshworks example?

00:02:27: Yeah tell him about that!

00:02:28: So he used Gemini for the logic and requirements right.

00:02:31: then he used FigmaMake to instantly generate the UX design structure And Then He Used Cursor To Write The Actual Code For The Prototype.

00:02:40: Right.

00:02:41: He built a functioning product roadmap application in two days.

00:02:44: Two days?

00:02:45: That is just... I mean, In normal enterprise environment Just getting the UX designer and lead engineer in same room for kickoff meeting takes two weeks

00:02:53: Exactly!

00:02:54: Which completely collapses that traditional pipeline we were talking about Boosh for Coos Cooner and Arundhav are pointing out.

00:02:59: this fundamentally moves the ceiling For what single-product manager can produce.

00:03:04: Yeah The era of writing specs And throwing them over the wall to engineering Is Over.

00:03:08: They're acting as actual builders now.

00:03:10: But I have to push back here for a second.

00:03:14: or at least ask the question, does this mean product managers need to learn to code?

00:03:19: Like are they all essentially becoming pseudo engineers now.

00:03:22: Well no not exactly!

00:03:23: They aren't writing code from scratch.

00:03:25: think of it like a movie director.

00:03:27: okay movie director.

00:03:28: yeah a director doesn't need to know how to wire complex lighting rig.

00:03:32: or you know sew costume?

00:03:33: Yeah...they just need to what the final scene should look and feel right.

00:03:38: in the past The Product Manager was a director describing a scene to a massive crew.

00:03:43: Today, they have this real-time CGI engine generating the entire scene as they speak —the completely new

00:03:50: discipline.".

00:03:50: They

00:03:50: are orchestrating that outcome rather than managing the assembly line—that makes sense!

00:03:56: But if one product manager can now do work of whole team in two days... The obvious risk is we just pump out useless features faster than ever, right?

00:04:04: Oh absolutely!

00:04:05: If the bottleneck isn't coding anymore it's just somewhere else.

00:04:08: Exactly and Brian Bonita has pointed something hilarious but painfully true about this he noted that engineering teams actually hate your AI prototypes.

00:04:19: Oh, I can completely imagine that.

00:04:21: Yeah he said if a PM just vibe codes something over the weekend meaning they prompt an AI to hack together A cool looking app without any thought for scalable architecture and drops it into slack on Monday morning The engineers just roll their eyes because

00:04:35: usually has the wrong tech stack zero connection of the company's data model And just completely lacks shared context

00:04:42: right?

00:04:42: Ai made the fast part of product development faster But it made the slow human parts infinitely more critical.

00:04:49: The bottleneck is now strategy and discovery, knowing what to build.

00:04:53: And that ties into what Diego Granados and Marine Kabud were saying.

00:04:57: They argue that AI's amazing at automating the mechanical busy work of a PRD Like

00:05:01: grouping hundred messy support tickets in themes.

00:05:04: Right?

00:05:04: Exactly!

00:05:05: And automating it finally frees up product manager To do hard thinking Because hard-thinking was never formatting document.

00:05:12: No, the hard thinking is answering the existential questions.

00:05:16: Why are we solving this specific problem?

00:05:18: Why right

00:05:19: now?".

00:05:19: Yeah and Marine made a brilliant point that AI forces you to audit your own weak.

00:05:25: You have to map out what is mechanical enough to automate And What actually requires Your strategic human judgment.

00:05:32: But here Is where We hit A really dangerous trap and Jeff got health issued A massive warning about This.

00:05:39: Regarding the discovery frameworks.

00:05:40: Yeah,

00:05:41: he was reacting to Pawe Heran who recently open sourced over a hundred AI skills for product managers basically encoding sixty-five different Discovery frameworks into prompts For tools like Claude

00:05:53: which is an incredible resource.

00:05:55: obviously

00:05:55: it Is but Jeff points out of vital flaw.

00:05:58: Frameworks only work because of the judgment you build while your struggling to fill them out.

00:06:02: That is such a crucial point.

00:06:03: If you spend three days synthesizing user interviews, your brain internalizes the nuances of

00:06:10: Exactly.

00:06:11: So if an AI agent instantly chains four of those skills together, it generates the questions synthesizes transcripts structures the opportunity tree in ten seconds.

00:06:20: The output looks incredible but you skipped the friction.

00:06:23: You bypassed this struggle

00:06:25: Right.

00:06:25: so my question is are we losing our cognitive calluses?

00:06:28: Are we just producing better-looking artifacts while making worse underlying decisions?

00:06:33: The risk is incredibly high...you can encode a framework into a prompt but you cannot encode the wisdom of knowing when to throw the framework away.

00:06:42: You end up with a very convincing hallucination of product strategy

00:06:47: and Steven Granny's offered, A Very Practical Reality Check here.

00:06:50: he argues that true discovery still requires human-to-human interaction.

00:06:53: Yeah?

00:06:53: You really can't outsource The reality

00:06:55: check?

00:06:55: No!

00:06:55: You Can't.

00:06:56: He says that underlying psychology of user interviews is key...you have To ask about past behavior not future wishes because

00:07:03: past behavior Is honest right future behaviors just aspirational

00:07:06: Exactly.

00:07:07: If you ask a user what they want, They will invent a feature.

00:07:09: that sounds great.

00:07:11: if you asked them What did last Tuesday when the system crashed?

00:07:14: You get actual evidence.

00:07:16: He really emphasizes asking about workarounds.

00:07:19: If a customer built a messy spreadsheet to work around A gap in your software That is real market opportunity.

00:07:27: AI cannot invent that grounded reality for.

00:07:30: So true.

00:07:31: Okay, so we're building prototypes at lightning speed.

00:07:33: We are trying to preserve our human judgments that we build the right things.

00:07:37: But who were we building these things for?

00:07:40: Well this is where the landscape just completely fractures.

00:07:43: The definition of a user is fundamentally changing.

00:07:46: Let's talk about Abhijit Khakhandiki his insight on this.

00:07:49: He said that to ten X your products adoption surface.

00:07:52: Right now you have to make it usable for AI agents.

00:07:54: Yeah

00:07:55: The next wave of software adoption isn't coming from humans clicking around a screen.

00:07:59: It's coming from machines autonomously interacting with your systems.

00:08:03: And Patrick Collison, from Stripe echoed this brilliantly.

00:08:06: He noted that Stripes developer experience their DX is increasingly being optimized specifically for agents

00:08:12: Because autonomous agents are actually hungrier For good DX than human developers Are.

00:08:17: Wait really?

00:08:18: Why Is That?

00:08:19: Well

00:08:19: think about it.

00:08:20: A Human User can look at a messy website Use Their Intuition and eventually find the hidden buy button.

00:08:28: An AI agent doesn't have human intuition.

00:08:31: Ah, right!

00:08:32: It relies on digital plumbing.

00:08:33: Exactly If your product does not have clear APIs or structured data The agent hits a brick wall.

00:08:40: You have to design machine-consumable workflows.

00:08:42: Celie Felipe broke this down into a really useful framework.

00:08:45: She mapped out the five types of AI in products.

00:08:48: Oh, I love this post.

00:08:50: Yeah she highlighted that when an autonomous agent becomes The actual user experience When the AI is taking actions on your behalf It completely changes all your assumptions about UX and economics.

00:09:01: Because

00:09:01: an Autonomous Agent doesn't just generate text it executes Actions that updates databases can send emails with very limited human supervision

00:09:08: right?

00:09:09: She noted that building an AI tutor That Just chats with a student is in a completely different universe of risk than an agent with access to a corporate credit card.

00:09:17: Which means the role of the product manager has to evolve again.

00:09:22: Mac Wallace argues that PMs are becoming AI orchestrators,

00:09:25: I like that term.

00:09:27: Yeah they're shaping complex decision pathways between humans and machines rather then just mapping linear user journeys.

00:09:34: But that raises a huge red flag for me regarding accountability.

00:09:37: When the machine is driving workflow across three different enterprise systems and it makes critical mistake, who owns this failure?

00:09:45: How do you even design trust when user isn't

00:09:48: human?".

00:09:49: That's defining challenge of the agentic era.

00:09:52: Accountability has to be designed into work flow itself.

00:09:55: You have to define exactly what system has confidence to act autonomously And what trip wires are forced to escalate back.

00:10:03: And

00:10:03: managing that level of complexity, it requires a completely different operating model at the macro organizational level.

00:10:10: Right?

00:10:10: Absolutely you cannot run these lightning fast autonomous products on an old corporate engine.

00:10:16: Multi-Schultz highlighted a glaring inefficiency here.

00:10:19: He noted that sixty five to seventy-five percent of product development time is currently spent purely on alignment and trying make decisions across departments.

00:10:27: Sixty Five to seventy five percent?

00:10:29: That's wild!

00:10:30: And yet most enterprise teams still manage their prioritization in disconnected spreadsheets,

00:10:35: which are just dead text in a grid.

00:10:38: A spreadsheet cell has no idea what your live customer feedback is saying it as no connection objectives, and key results.

00:10:45: Exactly!

00:10:46: You end up shipping the wrong things no matter how fast your AI helps you code them.

00:10:51: Antonia Lendi and Denise Tills argue that this is where a true product operations layer becomes non-negotiable.

00:10:58: But let's clarify what ProductOps actually is because lot of people just see it as middle management.

00:11:02: Right they argue It isn't an organizational janitor cleaning up Gira tickets...it is actual central nervous system connecting strategy to execution.

00:11:10: But setting that up completely fails if the leadership culture is broken.

00:11:15: Steven Archer called out what he sees as the biggest lie in enterprise development today, bolting Agile vocabulary onto a command and control IT structure.

00:11:22: Oh yeah calling it rigid deadline driven feature factory at two weeks sprint doesn't magically make your company agile.

00:11:30: Exactly!

00:11:31: Stephen's point is fundamental.

00:11:33: difference is trust.

00:11:35: Leaders need to provide the strategic context and trust empowered product teams.

00:11:42: Petra Will had an amazing analogy for this.

00:11:44: She said, a well-functioning product org isn't a military fire drill.

00:11:48: it is a flotilla of kayaks

00:11:50: A flotila?

00:11:51: Kayak's.

00:11:52: I love that image.

00:11:52: yeah

00:11:53: You have a fleet moving generally in the same direction But each individual boat is steered by someone who knows their specific stretch of water intimately you leave The micro adjustments to the edges of the organization

00:12:05: and pushing That judgment to the Edges Is the only way a company can survive.

00:12:08: the final layer we need To discuss today which is security.

00:12:12: Security in the age of AI feels incredibly daunting, honestly.

00:12:16: Kushpu Kashyap pointed out that Anthropics project mythos proves AI is completely collapsing the vulnerability window

00:12:23: Right because historically an enterprise had a buffer weeks or months between discovering a vulnerability and bad actors exploiting it.

00:12:31: now That window was down to days Or sometimes just hours

00:12:34: And the traditional backlog approach to security debt.

00:12:38: where you find a bug create a ticket fix it next quarter, that completely breaks down.

00:12:43: Cushpoon notes that AI tools will surface more vulnerabilities than any team has ever seen while adversaries use the exact same AI to exploit them

00:12:53: instantly.".

00:12:54: And harshest Ravatsa took this a step further regarding how we evaluate these massive AI models before they even go live.

00:13:01: He says evaluating multimodal AI can no longer be a subjective human test,

00:13:06: you Can't just look at the output and say yeah that looks generally correct.

00:13:08: No You can't.

00:13:09: he published an open evaluation operating system specifically designed to catch upstream failure modes.

00:13:15: because

00:13:15: when We talk about multi-model AI?

00:13:17: We're talking About models processing text images in audio all at once.

00:13:21: An upstream failure In one modality might Look like A bug in another

00:13:24: Right, like if the AI generates a misleading audio summary.

00:13:28: The root cause might actually be retrieval bias from the internal database or A weird text.

00:13:34: answer stems From the AI dropping an image caption entirely

00:13:37: and traditional security just checks for malware.

00:13:40: It completely misses these behavioral failures of the AI itself.

00:13:44: So my question is If product ops in Security have to act Like the central nervous system How do large enterprises build this without it devolving into just another slow layer of bureaucracy?

00:13:57: Well, the solution is embedding them directly in to the builder's workflow.

00:14:01: If your AI thinking partner is evaluating your code for modality bias while you are prototyping incursor It becomes an automatic reflex or

00:14:08: reflex instead a two-hour committee meeting

00:14:10: exactly.

00:14:11: we're orchestrating intelligence across the entire life cycle now.

00:14:14: Wow We have covered a massive amount of ground today, from PMs building apps in a single day to the friction of strategic judgment.

00:14:22: To designing for AI agents and rewiring The Corporate Nervous System.

00:14:26: That's

00:14:26: a lot to process for sure

00:14:27: It is.

00:14:28: Which leaves us with A final slightly provocative thought For you to chew on this week If autonomous AI agents truly become the primary users of tomorrow's enterprise software, will product managers in future even need to study human psychology?

00:14:44: Or they transition entirely into behavioral economists for machines.

00:14:47: That is a wild thought but very possible!

00:15:02: Thanks so much for joining us.

00:15:03: Thank you, and remember to subscribe!

00:15:05: We'll catch on the next deep dive.