Best of LinkedIn: Artificial Intelligence CW 15/ 16

00:00:00: This episode is provided by Thomas Alguyer and Franis, based on the most relevant LinkedIn posts about artificial intelligence from calendar weeks fifteen-and sixteen.

00:00:09: Franus supports ICT in tech providers with AI ecosystem strategy by delivering independent vendor assessment builds versus analysis an ecosystem intelligence that prevents expensive mistakes and positions the provider's competitively.

00:00:23: you can find more info description

00:00:26: And we are jumping right in.

00:00:27: Welcome to our deep dive, whether you're commuting or just tuning-in from your desk Our mission today is to unpack the absolute top AI trends and insights We've seen entirely from The LinkedIn community over the past two weeks.

00:00:40: Yeah, and we are keeping this super focused.

00:00:42: no fluff No basic definitions.

00:00:45: if you are professional on the ICT and tech industry This is exactly what you need To know about where the market Is actually heading

00:00:51: Right.

00:00:52: And to really understand where it's heading, we kind of have to start with a regulatory wake-up call that is hitting the industry right now because before we get into like all the cool autonomous AI stuff The governance reality is just fast approaching.

00:01:05: Oh!

00:01:05: It's right at the door.

00:01:06: I mean Tyrone May put out this post That Really Caught My Eye.

00:01:09: He gave this stark warning... ...that the EU AI Act officially kicks in on August second Which

00:01:15: is practically tomorrow in corporate

00:01:17: time Exactly and the fines go upto like fifteen million euros?

00:01:21: Yeah.

00:01:22: But the craziest part of his post was that most SME owners, small and medium businesses they basically have no idea.

00:01:30: They're even in scope?

00:01:31: yeah That's the scary part because you know people think well I'm not open AI iI am NOT training a massive language model so im fine.

00:01:38: but Michelle Bressoni highlighted this crucial detail.

00:01:41: You don't Even Have to build an AI To be liable

00:01:44: Right Because It's In Everything.

00:01:44: Now

00:01:45: Exactly.

00:01:46: Vendors are just embedding AI into email systems, your standard office suites support platforms and Bersony points out that the EU-AI Act labels you a deployer for using those systems.

00:01:58: Wait

00:01:58: so it's like renting an apartment?

00:01:59: And suddenly this city holds you legally responsible because the landlord installed smart thermostat.

00:02:07: That is actually a perfect analogy.

00:02:08: Yes,

00:02:09: that's wild.

00:02:09: Yeah And it's not even just the European problem.

00:02:12: Richard Scott posted this brilliant point about the global reach.

00:02:15: like if you are a us-based company and You use an AI tool to monitor the emails or workflows of an EU based employee.

00:02:24: boom

00:02:24: You're in scope.

00:02:25: Yeah, it instantly triggers the high risk legal obligations.

00:02:29: you are certainly playing by EU rules from Texas or California

00:02:32: Right and Armin K synthesized this really well.

00:02:34: he basically called the EU AI Act The new operating system for AI In Europe like its a foundational layer now.

00:02:42: But not everyone is thrilled about that Operating System.

00:02:44: Oh

00:02:44: definitely not.

00:02:45: Christoph Lucha brought up other side of his.

00:02:47: He sees as massive over regulation That could just kill investment.

00:02:51: You know, it's that classic tension between ethical guardrails and just heavy bureaucracy.

00:02:57: Yeah

00:02:57: It slows things down But honestly the compliance goes so much deeper than just EU-AI Act.

00:03:02: Look at what Paul Wiesbazarski posted about Dora regulations in banking sector.

00:03:06: Oh!

00:03:06: Dora?

00:03:07: The Digital Operational Resilience Act.

00:03:09: That is huge

00:03:10: Right.

00:03:11: Pazarsky pointed out this insane systemic risk.

00:03:14: If a bank relies on single AI providers say They just use open AI for everything and that provider goes down.

00:03:21: The bank's whole stack goes dark!

00:03:23: And door essentially makes that illegal.

00:03:25: now, you can't have a single point of failure.

00:03:27: Exactly so.

00:03:28: multi-provider routing is now mandatory fix.

00:03:30: You have to be able instantly switch to another AI if one crashes

00:03:34: which is a complete architectural nightmare to build.

00:03:38: And I mean, this governance challenge is accelerating specifically because the tech itself is changing so fast.

00:03:44: we aren't just dealing with chat windows anymore

00:03:46: right?

00:03:46: The whole paradigm is shifting

00:03:47: exactly!

00:03:48: We're moving to agintic workflows Autonomous Execution.

00:03:51: Luisa Ligioni had this great post where she observed a boardroom meeting.

00:03:56: The execs were all bragging that eighty-eight percent of their company uses AI tools.

00:04:00: Love you guests, just generating emails and summarizing meetings.

00:04:03: Bingo!

00:04:04: She pointed out.

00:04:05: almost zero per cent are actually using AI agents.

00:04:08: You know systems execute work autonomously without a human sitting there prompting it step by step.

00:04:13: Yeah but the gap is closing fast on the dev side.

00:04:16: Eduardo Fernandez shared this update about VS Code Agent's insiders And completely flips script for software development.

00:04:23: Oh Microsoft Update.

00:04:25: So previously AI was kind of a passenger in the IDE, right?

00:04:29: You ask it to question and suggest some code.

00:04:31: But now the AI is literally the driver!

00:04:34: It handles isolated work trees...it runs the terminal…it hands things off to

00:04:39: cloud.".

00:04:39: It's writing & testing its own work.

00:04:41: Greg Coquillo from Breakish or Panday had fantastic insights on this.

00:04:45: They were explaining that systems like Claude Code are acting as actual operating system for software execution now.

00:04:53: Like an OS just for the AI?

00:04:55: Yeah, complete with sub-agents.

00:04:57: So the AI has its own skill registries and these master loops where it constantly evaluates its own progress.

00:05:02: It's not just generating text...it's managing a project.

00:05:05: See that sounds incredibly complex And honestly it sound super locked into the cloud like what if you want to own it?

00:05:11: What If You Don't Want To Send All Your Proprietary Code To Anthropic Or Microsoft?

00:05:15: That's

00:05:15: The Big Enterprise Hesitation Right There.

00:05:17: Well,

00:05:17: Shubham Sabu actually posted the answer to that.

00:05:19: He shared an update on The Hermes Agent and this is huge!

00:05:22: It runs fully local...

00:05:24: Wait a hundred percent local?

00:05:25: ...a hundred-percent

00:05:25: free, fully local And it's self improving via Elama.

00:05:30: So just run two hundred four seven on your own hardware doing tasks and it never sends A single packet of data into cloud.

00:05:36: Okay see That changes everything because when you bring these powerful autonomous agents into the enterprise, but you keep

00:05:53: More ROI.

00:05:54: Exactly, they went from saying hey show me which NEI can do to just demanding how fast can I get costs out of my business?

00:06:00: and where does this hit my P&L?

00:06:02: Yeah the honey mean phase is over And Magdalena Johns-Fack & Jack Infectioner really dug into why that is.

00:06:07: They argue that AI adoption is failing at scale right now.

00:06:10: Failing

00:06:11: Really with all these capabilities.

00:06:13: yeah But not because the models are weak.

00:06:15: it's because a fragmented architecture in governance.

00:06:18: Companies are treating AI like these cute little isolated pilots.

00:06:23: Fettner says it needs to be treated as a core platform capability.

00:06:26: Okay, but I have to push back a bit here.

00:06:29: if these models are literally doing PhD level reasoning Why our massive enterprise is still tripping over the starting line?

00:06:36: Well Dr Sheik and Isaacs posted The Harsh Truth on that.

00:06:39: he said AI Is A Quality Multiplier.

00:06:41: It Is Not A Fix.

00:06:43: So If Your Data Is Trash

00:06:44: Exactly!

00:06:45: If your data is garbage the AI doesn't fix.

00:06:47: it just generates systemic errors at high speed.

00:06:50: its scales.

00:06:50: you're mess.

00:06:51: That makes so much sense.

00:06:52: And it's probably why we're seeing entirely different strategies globally, like Sumshoe Bro Pell Chudry shared this really fascinating macro strategy view

00:07:00: about the India playbook?

00:07:02: Yeah So The US is doing this super capital heavy build AGI at all costs thing.

00:07:06: China is doing stake driven scale but india Is building what he calls a third Playbook

00:07:11: the frugal approach

00:07:12: right.

00:07:13: It's an efficiency led model focused on small language models or SLMs.

00:07:18: They're building enterprise services under real world compute constraints rather than just burning billions on GPUs.

00:07:26: It's smart, but you know as we scale these efficiency models and deploy all of the autonomous agents We are introducing entirely new categories risk risks that go way beyond standard IT security.

00:07:39: Yeah letting an agent run autonomously is terrifying.

00:07:44: Romy Alcafaje and Ossoff Nakash brought up Project Glasswing.

00:07:48: Have you heard of this?

00:07:49: I saw the name floating around, yeah.

00:07:51: So

00:07:51: Claude Mithos has been actively finding decades-old zero day vulnerabilities in critical software Stuff humans have missed for years

00:07:58: Though it's like a super powered security guard

00:08:00: It is.

00:08:00: but as Nakosh points out its starting to look nuclear capability.

00:08:04: Because if an AI can find a Zero Data Patchit A malicious AI could find it to exploit it Coordination becoming way more important than just local control.

00:08:12: That is slightly terrifying for anyone in IT.

00:08:16: But to bring it back down the earth, Rob Vanderveer offered a really practical solution.

00:08:20: He shared the OWASP AI threat modeling.

00:08:22: one pager.

00:08:23: Oh nice something actionable

00:08:25: Super actionable.

00:08:26: It helps companies quickly identify their relevant Ai security risks without getting totally paralyzed by you know The existential dread of autonomous hackers.

00:08:35: Yeah we need those frameworks Because there's also a much more subtle kind of risk creeping in.

00:08:40: Borger posted about this.

00:08:42: regarding Claude Opus, four point seven...

00:08:44: Okay what do they find?

00:08:45: So the AI now uses adaptive reasoning.

00:08:48: basically it decides on its own how hard to think about your prompt.

00:08:52: Wait!

00:08:52: It judges the prompt Exactly.

00:08:54: and because these models are heavily benchmarked on coding and math it saves its really deep heavy reasoning for software engineering tasks.

00:09:02: but if you ask to do strategic business work like writing a restructuring plan, it routes it as low effort.

00:09:08: Wait

00:09:09: so its like hiring a literal genius consultant who only pays attention if you ask them the math question but completely zones out when you asked him to fix your company's business model?

00:09:18: That is exactly what itís doing!

00:09:19: Thatís insane!

00:09:21: Itís completely short changes like a billion knowledge workers that arenít developers

00:09:25: Right.

00:09:26: and this why Guy Jopmo and Marco Geier concluded that the ultimate competitive advantage isn't the AI alone.

00:09:33: It's The Human in the Loop!

00:09:34: Yes, human-machine collaboration it is about assigning the right tasks to the right intelligence And just refusing let machine dictate value of human strategic knowledge.

00:09:45: That is such a critical take away which brings me into question.

00:09:48: I want leave you our listener with today.

00:09:51: Think About Your Own Workflows.

00:09:53: Are you just buying AI tools and hoping a strategy magically appears, or are you actively redesigning your workflows in governance so that you control the AI rather than other way around?

00:10:03: That is the million dollar question right now.

00:10:05: Literally!

00:10:06: And hey if you enjoyed this episode new episodes drop every two weeks.

00:10:09: also check out our other editions on ICT & Tech digital products and services cloud sustainability and green ICT defense tech and health tech.

00:10:18: Thank

00:10:18: You So Much for Joining us For This Deep Dive.

00:10:20: Make Sure To Hit Subscribe.

00:10:21: We'll Catch You Next Time.

00:10:22: Thanks everyone.

00:10:23: See you on the next one!