The Technology Works. The Deployment Is What Fails

Episode Summary

Tallulah Le Merle spent most of the last decade on digital and data transformation with large corporates before moving to the investment side at Fifth Era, and describes the same failure repeating with AI. A leader hears innovate or die, goes to a conference, gets excited about a tool, and licenses it for a year or so. What follows is a scatter gun deployment that never started from strategy, and an executive who has spent heavily and cannot see the dial move. Her alternative is to flip the equation: start from the needs of customers and employees, find where the organization is leaking value, get specific about use cases one value pool at a time, and only then look at tools. She is relaxed about being late, arguing that fast follower is a legitimate position when innovation is moving at this pace. The second half turns to the workforce, the emotional use cases, and what any savings are for.

Key takeaways

  • The failure is deployment, not technology. She cites an MIT report finding that 90 percent of organizations are not getting the return they expected, and reads it as a problem with how the technology was deployed rather than with the technology or its promise

  • Flip technology first into needs back. Start from the needs and pain points of customers and employees, find where the organization is leaking value or running slow manual processes, and work from those value pools to use cases and only then to tool

  • It is fine to be a fast follower. With innovation moving at this pace there is no requirement to be a first mover, and looking at what competitors or similar organizations in other industries already use is a legitimate strategy

  • Diligence the tools properly, because a lot of them are wrappers. There are at least ten teams building for any given enterprise use case, so time spent comparing solutions against a specific need is time well spent rather than a delay

  • Change management is the part nobody wants to do. People focus on the platform and the software rather than the capabilities uplift, on helping employees understand what it means for their day-to-day role and how they are incentivized to grow within it

  • Nominate champions and give them the job. Find the rising stars who are excited to get hands-on, make them responsible for being a hybrid human and AI employee, for making the solutions fit the company, and for teaching everyone else

  • This cannot be delegated to IT. Her argument is a chief AI officer at the ExCo level reporting to the CEO, because it has to be cross-functional and thought through from the top rather than solved bottom up inside silos
    Cost is not the only prize. She puts most enterprise adoption right now at efficiency and productivity, and argues organizations should also be looking at top-line growth, new business and revenue models, and getting to market faster out of slow multi-month R&D

  • Everyone underestimates inertia. She cites a Goldman Sachs report putting the peak unemployment effect at 0.5 points, against five in the last recession and around eleven in COVID, and reads the shift as gradual and mostly about augmentation

  • The emotional use cases need people who are not necessarily technologists. Her argument is that the best placed people to build for those needs are the ones who understand them: creatives, caregivers, right-brainers, social scientists, psychologists and neurodivergent people

About Tallulah Le Merle

Tallulah Le Merle is a partner at Fifth Era, working on the investment side of AI, and describes most of the preceding decade as time spent with large corporates and enterprises on digital and data transformation and on AI strategy back when it was machine learning and big data. She is also a fractional executive and a consultant, and her platform sits at the intersection of AI and humanity, where she argues what she calls the case for hope: lucid about the risks of safety, security, data privacy, cognitive offloading and bias, and unwilling to concede the trajectory to them. On the firm, Fifth Era takes its name from the fifth era of human evolution, after hunter-gatherers, agrarians, industrialists and the information era. She spent twelve years in the UK and France and is now back in the Silicon Valley Bay Area, where she grew up, and describes her own edge as bridging the world of large corporates with the world of fast-paced innovation.

 

In this episode

00:00 Welcome and guest introductions
01:45 Ken: GDPR, privacy, and the road to AI governance
03:32 Bob: when the models changed and the controls did not fire
04:50 What governance as code actually looks like
05:54 Where the policy code lives
07:06 Why gates also have to run at runtime
08:21 Will the CI/CD vendors build this?
09:20 Why the tooling is open source
11:07 Agent swarms, or viruses with credit cards
11:52 Ford: the puddle that flipped the car
14:10 GM: governance treated as a safety system
15:14 Over-the-air updates and automated targeting
16:07 Governance After Hours in San Francisco
16:42 The biggest misconception: governance as a brake
17:34 Unsafe at any speed
18:00 How much testing is enough
18:24 Red teaming and adversarial testing
19:14 The security analogy: shift left
19:53 Getting past the maybe gate
20:33 How many models do you test against
21:05 Inside Beacon
22:19 Umbrella, Lantern, and the audit layer
23:14 The Lean AI Handbook and the learning loop
24:47 Blast radius control and rollbacks
25:27 Data as the new oil, refined
26:54 Bob on where to start
28:23 Ken on his free e-book
29:26 Executive clarity and prototype theater
30:33 The one thing to remember
31:26 Wrap-up

 In Tallulah’s Word

“It’s not a problem with the technology and the promise of the technology, it’s a problem with the deployment of it”

— Tallulah Le Merle   (08:14)

“Instead of being technology first, it’s about being people back or needs back”

— Tallulah Le Merle   (05:35)

“It’s okay to be a fast follower. You don’t need to be a first mover in adopting tech”

— Tallulah Le Merle   (07:00)

“But the truth is, everyone underestimates inertia. This shift will be much more gradual than what we’re hearing.”

— Tallulah Le Merle   (15:35)

“This has to be cross-functional. It has to be thought of right from the top and then trickle down across functions”

— Tallulah Le Merle   (13:18)

“The best placed people to build for those use cases are not necessarily technologists, deeply technical experts, et cetera, but it’s the people who understand the needs”

— Tallulah Le Merle   (21:17)

 

Resources

Tallulah Le Merle and Fifth Era

  • Tallulah Le Merle on LinkedIn: linkedin.com/in/tallulahlemerle. The one-stop shop she names on air, which she says carries links to her website and podcast

  • Fifth Era: The firm where she is a partner on the AI investment side. Named for the fifth era of human evolution, after hunter-gatherers, agrarians, industrialists and the information era

  • Her daily AI news show: Described on air as the first fully AI-generated news show about AI, 15 minutes or less, delivered by her AI twin. Wednesdays are Workforce Wednesday, covering enterprise AI and what some of the key organizations are doing

Ideas and frameworks discussed

  • Needs back, not technology first: Her core reframe. Start from the needs and pain points of customers and employees, identify the value pools where the organization is leaking value, get specific about use cases inside one of them, and only then evaluate tools

  • The fast follower position: Permission to not move first. With at least ten teams building for any enterprise use case and innovation at this pace, watching what comparable organizations adopt and trying it once it is proven is a strategy rather than a failure of nerve

  • Champions rather than mandates: Nominating rising stars in each function, making them responsible for being a hybrid human and AI employee, for making the solution fit the company, and for teaching the rest of the organization

  • A chief AI officer at ExCo level: Her structural answer. Reporting to the CEO, cross-functional, working top down across functions and operating companies, rather than pushed into IT or left to individual silos

  • Cost out versus top line: Most adoption right now aims at efficiency and productivity. She argues organizations should also be looking at new business and revenue models, serving customers in new ways, and cutting multi-month R&D cycles

  • Where the savings go: The question of whether efficiency savings are reinvested in people, markets and products or passed to shareholders. She says it is too early to generalize, that the answer will differ by organization, and that reinvestment is how augmented roles get reabsorbed

  • The emotional use cases: Her argument that the needs surfacing now are best understood by creatives, caregivers, right-brainers, social scientists, psychologists and neurodivergent people, and that whole product categories and industries may arise there

  • The case for hope: Her platform at the intersection of AI and humanity. Lucid about safety, security, data privacy, cognitive offloading and bias, and built on the argument that the people who care about those risks have to stay in the conversation rather than leave it

Named on air, as she cited them in October 2025

  • An MIT report: Cited for the finding that 90 percent of organizations are not getting the return they expected from their AI investments

  • The McKinsey State of AI Report: Cited for 80 percent of organizations adopting AI at scale in three or more functions, which she reads as the end of AI as an experiment

  • A Goldman Sachs report: Cited for a peak unemployment effect of 0.5 points, which she sets against five points in the last recession and around eleven in COVID

  • A World Economic Forum report: Cited for roughly 70 million jobs that could be replaced by 2030 against around 160 million new jobs created

  • The GPT-5 reception: Her example of the emotional use cases surfacing. The complaint was that it did not have the same tone, texture and feel of interacting that 4 had, which she reads as evidence people were using the technology to feel seen and heard

  • AI Realized Summit: She moderated a panel at the summit on 5 November 2025, a few weeks after this conversation was recorded

Related AI Realized episodes and events

 

Frequently Asked Questions

 
 
 
 
 
 
 
 
 
 
 
Previous
Previous

Shadow AI Is a Permission Problem, Not a Tool Problem

Next
Next

Zero to Campaign With Everyday AI, in Four Steps