Best of LinkedIn: Artificial Intelligence CW 07/ 08

00:00:00: This episode is provided by Thomas Allgaier and Frennus, based on the most relevant LinkedIn posts about artificial intelligence from calendar weeks seven and eight.

00:00:08: Frenness supports enterprises with market-and competitive intelligence decoding emerging technologies customer insights regulatory shifts and competitor strategies.

00:00:18: so product teams and strategy leaders don't just react but shape the future of AI.

00:00:23: Welcome to The Deep Dive everyone!

00:00:24: We are looking at top Artificial Intelligence trends surfacing across LinkedIn.

00:00:29: Yeah, specifically we're pulling insights from calendar week seven and eight of twenty-twenty six.

00:00:34: And looking at this massive stack of research today I really have to say the vibe has shifted it

00:00:40: Really?

00:00:40: Has that initial wow faggot of generative AI is well basically fading out entirely

00:00:46: right?

00:00:46: It feels like Twenty twenty three was The Wild Party and now the lights are on the music Is off and the manager's here To check the books.

00:00:53: That is a great way to put it.

00:00:54: you're picking up On This huge shift toward operating discipline.

00:00:58: We just aren't talking about magic tricks anymore.

00:01:00: The conversation has moved from whether you should adopt AI to how you actually make it compliant economically sustainable and embedded in real workflows

00:01:09: without going broke or getting sued In the process.

00:01:12: exactly

00:01:12: so for you listening today, we're going to cluster these insights into three main themes So you can make sense of at all.

00:01:19: First will hit a I governance and sovereignty.

00:01:22: Then we'll look at enterprise execution and reality checks, And finally will unpack the technical shift toward agents in architecture.

00:01:29: Let's

00:01:29: jump straight into governance because it is no longer just about high-level ethics theory.

00:01:34: It's really about market survival at this point.

00:01:36: Yeah David Pant had a post that honestly kind of set the tone.

00:01:40: he calls The situation a ticking time bomb

00:01:42: which is incredibly accurate when you look at the calendar.

00:01:46: He is pointing straight at the EU AI Act deadlines that hit in August, twenty-twenty six.

00:01:52: That's less than seven months away from this time of these posts!

00:01:56: It isn't theoretical anymore...

00:01:58: No it's imminent and Pat makes very clear noncompliance not just a slap on his wrist.

00:02:04: we are looking for fines up to thirty five million euros or and this is the terrifying part, seven percent of global revenue.

00:02:12: Seven percent of revenue?

00:02:13: Yeah!

00:02:13: Not profit.

00:02:14: that's kind of number keeps a CFO awake at night plus you risk getting immediately banned from European markets.

00:02:20: And The Kicker here which Pant specifically notes Is there no grandfathering?

00:02:26: Right

00:02:26: You can't just wave legacy pass.

00:02:27: Exactly If you deployed system two years ago still have to retrofit it meet these new standards you have to open up those black box algorithms and prove they are safe.

00:02:37: Which perfectly brings up Patrick Sullivan's observation, he pointed out that U.S.

00:02:41: executives are reading these EU rules in just getting totally confused...

00:02:44: Because they're looking for the word ethics?

00:02:47: Yes!

00:02:48: They treat it like a corporate philosophy exercise.

00:02:50: but Sullivan says no the EU-AI Act is written in the language of product safety.

00:02:54: Its FDA approvals its conformity assessments.

00:02:57: Exactly It's like crash testing car.

00:02:59: not debating Asimov laws of robotics

00:03:02: And at Product Safety angle leads into a really fascinating strategic pivot proposed by Marco Van Herne.

00:03:09: He argues that Europe shouldn't even be trying to compete on building the foundational models

00:03:14: because The US and China have basically won that specific race,

00:03:17: right?

00:03:17: So van herne says europe should focus entirely on scaffolding.

00:03:21: I Really love that concept.

00:03:22: when i read it the scaffolding is like the cage around the model.

00:03:25: That's the perfect visual.

00:03:26: the AI model itself Is a wild unpredictable engine.

00:03:31: The scaffolding is a trust infrastructure, it's the audit trails, the stripped guardrails that compliance reporting.

00:03:37: Van Heern argues Europe can actually build and export this trust infrastructure as a premium product.

00:03:43: so the US sells the chaotic engine in Europe cells breaks and steering wheel.

00:03:48: And In heavily regulated global market those brakes are incredibly valuable.

00:03:52: you cant drive car on road without them.

00:03:55: We're already seeing play out specific local laws too.

00:03:58: Jungan highlighted Italy's new law number one.

00:04:01: thirty two.

00:04:02: Oh, the workplace law?

00:04:03: Yeah it is incredibly strict.

00:04:05: if you are using AI for anything related to recruitment task allocation or performance evaluation It triggers massive transparency obligations.

00:04:14: You

00:04:15: absolutely cannot just let an algorithm filter resumes or fire a gig worker in Italy anymore, you need proven human oversight!

00:04:21: Which introduces huge data sovereignty issue right?

00:04:24: Yulia Habriel brought up really alarming point about the US Cloud Act.

00:04:28: In relation to all this compliance

00:04:30: This is major Sovereignty Trap.

00:04:32: The cloud act basically means that if your using a us based vendor

00:04:35: Even If servers are physically sitting in Frankfurt or Paris

00:04:38: Right If the vendor is an American company, US authorities can legally demand access to your sensitive corporate data.

00:04:45: So a European company using a U.S.

00:04:47: platform to manage its European compliance documentation is still subject to U. S law?

00:04:52: It defeats the purpose of local sovereignty and it's causing some companies to take incredibly drastic measures.

00:04:58: Coon DeWitt from lead fabric literally banned open AI products entirely from his company

00:05:04: which sounds kind of wild for a tech company in twenty twenty six.

00:05:07: It sounds like leadism at first, but his reasoning is pure strategic sovereignty.

00:05:12: He isn't saying the tech as bad.

00:05:14: he has prioritizing ethical practices and absolute infrastructure

00:05:18: independence.".

00:05:19: He specifically cited OpenAI's erratic policy changes in their growth-at-all costs mindset didn't he?

00:05:25: Yes!

00:05:26: He basically said we don't want to build our business on rented land where the landlord might randomly change the locks...

00:05:33: That makes total sense.

00:05:34: governance isn't just paperwork.

00:05:35: it is becoming a core competitive strategy.

00:05:38: But let's shift gears to our second theme, which is enterprise execution and reality checks because even if your scaffolding is perfectly compliant you still have make the tech actually work

00:05:49: And data shows that most companies are failing at that.

00:05:51: Ashish Doan called this great reckoning.

00:05:54: The numbers he shared were just brutal.

00:05:55: Big Tech committed something like six hundred thirty billion dollars to AI infrastructure

00:06:00: but when look actual enterprise pilots out in field They are failing at a ninety-five percent rate.

00:06:06: Ninety five percent, if any other department had a ninety five percent failure rate the entire leadership team would be fired.

00:06:12: why is it so high?

00:06:13: A lot of comes down to complete lack financial discipline.

00:06:17: Mark Byershoder gave great breakdown this.

00:06:19: he says.

00:06:20: leaders still governing AI spending as if were traditional software subscription

00:06:25: Like you pay flat fee for C and just use that much exactly.

00:06:29: but AI doesn't work like.

00:06:31: It is utility consumption, it runs on tokens.

00:06:35: Byershoter warns about this invisible consumption problem

00:06:38: Right where every prompt spins the meter.

00:06:40: Yes

00:06:41: If your team using a massive expensive reasoning model just to summarize basic PDF or write an email you're burning cash.

00:06:49: He estimates companies are inflating their spend by five-to ten times.

00:06:53: Just use wrong models for simple tasks.

00:06:56: It's like chartering private jet just go pick up groceries.

00:06:59: Yes, it gets you there but the economics are totally broken.

00:07:02: And...it is not just budgets feeling this strain It's employees too.

00:07:06: Hina Pirrohit shared some really interesting research on the whole productivity miracle narrative.

00:07:11: We always hear that AI makes us so much faster!

00:07:14: It does make individual tasks faster But her data shows its actually leading to severe workload creep.

00:07:19: I have absolutely felt this.

00:07:21: The AI drafts the report in five minutes, so suddenly my boss expects me to do ten times the volume of reports!

00:07:27: Exactly employees are taking on much broader tasks.

00:07:30: now you have product managers trying to write code and because prompting an AI feels like chatting...the boundaries Totally blurred.

00:07:38: It doesn't feel like you're doing heavy lifting, but the cognitive strain of constantly multitasking and context switching is huge

00:07:44: right?

00:07:44: You are doing a lot more But you're not necessarily doing better work

00:07:48: which brings us to the organizational chaos.

00:07:52: Eduardo Ordex had a hilarious but painfully accurate Take on corporate structures Right now.

00:07:59: he pointed out the absurdity of these massive AI task forces.

00:08:02: yeah

00:08:03: Companies are rushing to appoint chief AI officers and assemble these grand ethics committees, but when you look at who is actually shipping the prompts?

00:08:11: And doing the work.

00:08:11: It's Jack the intern

00:08:13: exactly it's The most junior person in the room copy pasting sensitive data into a chat window.

00:08:19: We don't need more high-level Committees.

00:08:21: we need to empower the actual doers with proper training.

00:08:24: Okay, we need a glimmer of hope here.

00:08:26: Is anyone Actually executing this well?

00:08:29: yes Ali K. Miller shared an incredible success story from Ginggo Bioworks.

00:08:34: They managed to achieve a forty percent cost reduction in protein production.

00:08:37: And they definitely didn't do that just by deploying a chatbot

00:08:40: Not at all, they built a fully integrated loop.

00:08:43: AI proposed the experiments Then actual physical robots ran those experiments in lab.

00:08:49: The resulting data was fed right back into GPT-V and human scientists oversaw & updated protocols.

00:08:56: So it was a true continuous loop of AI, robotics and humans?

00:09:00: Yes.

00:09:01: And it perfectly validates what Magdalena Picciariello has been advising.

00:09:05: She says companies need to stop using AI just to generate text... ...and start using it.

00:09:09: simulate futures.

00:09:10: Simulate futures I like that phrasing

00:09:12: Instead asking the AI.

00:09:13: to summarize meeting ask you build predictive model.

00:09:16: What happens our supply chain if shipping costs spiked by two percent?

00:09:20: That predictive simulation is where massive return on investment lives.

00:09:24: But to run complex simulations and physical robotics loops, you can't just use a simple web interface.

00:09:30: You need a completely different architecture which brings us to our third and final theme agents an architecture.

00:09:38: This is where the technology's fundamentally shifting.

00:09:40: Xavier Agnetti declared that The Rapper Era Is Completely Dead

00:09:45: Meaning those startups That Just Built A Pretty User Interface Wrapped Around GPT-IV Are Done

00:09:50: Finished.

00:09:51: Agnetti says the future is building an enterprise AI operating system.

00:09:55: So what does that new tech stack actually look like?

00:09:58: for you listening?

00:09:59: Bridge Keyshore Panday outlined a really clear four-layer architecture for this.

00:10:03: It's great roadmap.

00:10:04: layer one is RA retrieval augmented generation.

00:10:08: that is basically table stakes now.

00:10:10: It grounds the AI in your own company data, so it gives factual answers...

00:10:13: Right and Layer Two?

00:10:14: Layer two introduces agents.

00:10:16: these aren't just chatbots.

00:10:17: they are systems designed to do work.

00:10:19: They can plan a sequence of tasks observe the results and act on them autonomously.

00:10:22: Okay layer three is one we're seeing everywhere.

00:10:25: right know MCP The model context protocol.

00:10:28: help us demystify That.

00:10:29: think of mcp as the universal USB port for artificial intelligence.

00:10:34: Before this protocol, if you wanted your AI to talk to your CRM system... ...you had to build a custom wire.

00:10:40: MCP is the standardized plumbing that lets the AI access your enterprise tools instantly and securely.

00:10:46: That makes it so much more scalable.

00:10:48: In The Final Layer, layer four A-to-A

00:10:51: Agent To Agent Communication This is where your specialized sales agent automatically coordinates with your specialized legal agent to draft and approve a contract without a human constantly playing middleman.

00:11:02: It sounds incredibly efficient, but handing over that much autonomy introduces some scary new risks.

00:11:08: Anup V. Prabhakaran warned about phenomenon he calls context blindness.

00:11:12: He shared a story about a customer named Sarah.

00:11:14: that perfectly illustrates why this is so dangerous.

00:11:17: Yeah, let's walk through that.

00:11:18: imagine an AI agent working autonomously at bank.

00:11:21: it is monitoring accounts and sees that a customer, Sarah suddenly stops depositing savings.

00:11:26: simultaneously her credit card spending spikes by fifteen thousand dollars.

00:11:30: So the AI looks to data does the math And concludes she has high risk.

00:11:36: It automatically cuts our credit limit to protect the Bank.

00:11:38: The Math Is Right But the context that AI entirely missed is that Sarah is in middle of a family medical emergency.

00:11:46: And crucially, she holds two million dollars and long-term investments at exact same bank.

00:11:51: So the AI operating this narrow blind loop just alienated high net worth client during worst week her life?

00:11:58: It's catastrophic business error made it lightning speed!

00:12:02: Which is why Ramana Roth argues as we move into agent era... Speck quality is your new competitive mode.

00:12:08: He used the term technical deflation, right?

00:12:10: Yes If AI can write the actual code for practically zero cost The value of just being a coder deflates.

00:12:17: The real values shifts entirely to humans who architect systems To find context and verify outcomes.

00:12:24: If specifications are bad Your agents will make decisions much faster.

00:12:28: But capabilities of these tools are accelerating so fast.

00:12:32: Armand Ruiz pointed out Manus Agents running directly on telegram.

00:12:35: You can just chat in your messaging app and have an agent execute complex research reports.

00:12:40: And Wieland Holfeldt noted that Gemini three point one pro is achieving massive, measurable jumps in abstract reasoning.

00:12:48: The agents are getting sharper by the week.

00:12:50: they really are.

00:12:51: but to wrap up this technical section I want to pivot to a really sobering financial warning from Julian Simon.

00:12:57: We talked about enterprise budgets earlier, but Simon is looking at the macroeconomics of.

00:13:10: Companies are taking out multi-billion dollar loans to buy these H. one hundred and B two hundred chips,

00:13:15: And they're using the hardware itself as a collateral for the loans like a mortgage on house

00:13:20: exactly.

00:13:21: but unlike real estate which generally appreciates Simon points out that GPUs lose seventy to eighty percent of their value in just two years because The next generation is always so much faster.

00:13:33: But

00:13:33: the banks writing these loans are assuming that chips will retain something like fifty percent of their value.

00:13:38: Right, so you have a trillion dollar infrastructure boom backed by collateral That is rapidly turning into obsolete scrap metal.

00:13:45: If the AI software revenue doesn't materialize fast enough to pay off the debt that entire financial structure could collapse.

00:13:51: Well,

00:13:52: it's a really sobering reality check.

00:13:54: to end on we have European regulators at the door ninety-five percent of enterprise pilots failing and A mountain of depreciating debt sitting in the server rooms.

00:14:04: It sounds bleak when you list it all out but this is exactly what happens When a hype cycle finally matures into a real industry The focus on governance and real architecture?

00:14:14: Its painful But its totally necessary.

00:14:17: We are finally treating AI like a business instead of a toy.

00:14:20: Exactly!

00:14:21: So before we sign off, I want to leave you with one final thought to mull over...we talked about European scaffolding and we talk about autonomous A-to-A agent to agent communication.

00:14:30: Imagine the scenario where an Autonomous Agent from U.S.

00:14:33: Enterprise tries to dynamically negotiate a supply chain contract With an Autonomous Agent From a European Enterprise.

00:14:41: Oh wow Both are operating autonomously, but they're running on entirely different compliance staff-holding.

00:14:48: Who audits the AI?

00:14:49: that audits The AI when foundational regulatory frameworks literally contradict each other?

00:14:55: That is going to be an absolute legal battleground very soon!

00:14:58: Exactly it's something you keep in mind as build out your own enterprise systems

00:15:03: And that wraps up our deep dive into weeks seven and eight.

00:15:06: If you enjoyed this episode new episodes drop every two weeks.

00:15:10: Also check out our other editions on ICT and Tech, Digital Products & Services, Cloud Sustainability in Green ICT, DefenseTech And HealthTech.

00:15:19: Thank you so much for your time today.

00:15:20: keep asking the hard questions about your AI strategy.

00:15:22: absolutely.

00:15:23: don't forget to subscribe.

00:15:24: see never risk a deep dive into changing world of tech.

00:15:27: See ya next time!