What happened: At the inauguration of its Seoul office this week, Anthropic simultaneously announced three colossal deployments at South Korea's conglomerates (chaebols). NAVER, South Korea's internet leader, is rolling out Claude Code across its entire engineering organisation. Samsung SDS is deploying Claude Cowork and Claude Code at Samsung Electronics — which employs more than 280,000 people. And LG CNS is extending Claude access to every employee of the LG Group. Three groups, nearly 700,000 employees combined, all switched over in a single announcement.

The Codito analysis: This series marks a category shift for enterprise AI. For three years, large-scale deployments mostly involved players in tech, consulting or financial services (KPMG, PwC, JPMorgan…). What Korea signed this week is the entry of heavy industry: consumer electronics, semiconductors, chemicals, telecoms, mobility, batteries. The logic changes. When a manufacturer like Samsung rolls out Claude Code to its thousands of engineers, it's no longer to "test AI" — it's because in-house software production (firmware, drivers, business tools, customer platforms) is becoming an industrial competitiveness advantage. And this Korean move precedes a European one already getting under way: Schneider, Saint-Gobain and Stellantis have started. Your clients and your competitors are on their way there.

What it means for you: For the head of a French SME, the useful reading isn't in the volume — you won't have 280,000 employees to equip. It's in the decision-making discipline that set these groups in motion. Three principles stand out from their announcements: (1) one targeted use case per function (engineering at NAVER, dev at Samsung SDS, support at LG), (2) a rollout in measured cohorts before going company-wide, (3) integration into existing tools — no parallel stack. Three principles that transfer to 50 or 500 employees without distortion. That's exactly what Part 2 is about.

The number of the week Nearly 700,000 employees switched over simultaneously. That's the combined headcount of NAVER, Samsung Electronics and LG Group gaining access to Claude this week. For a sense of scale: it's more than the entire working population of Brittany. And it's the equivalent — combined — of several KPMG-style announcements in a single week. The pace of industrial deployments is accelerating, and East Asia is now setting the tempo.

At first glance, the idea that Samsung or NAVER announcements could enlighten a French SME seems far-fetched. And yet, behind the spectacular volumes lie three disciplines that transfer wholesale to 30, 100 or 500 employees. Here they are.

1. One use case per business function, not one use case per technology

Samsung didn't deploy Claude "for productivity" — it deployed it at Samsung SDS on precise technical use cases (code review, test generation, automatic documentation). NAVER did it for engineering. LG for cross-functional support roles. Every time: an identified business function, a measurable case, a quantified before/after. For an SME, the transposition is simple: before signing the slightest POC, identify two or three recurring tasks (sales writing, qualification, client summaries) where you can quantify the time saved. Without that framing, AI remains a collective gadget. With it, ROI can be measured from week one.

2. Rollout by cohorts, not by decree

None of these groups deployed in one block. NAVER started with its cloud-infrastructure teams before extending to the app side. Samsung SDS began with the teams most fluent in dev tools. LG piloted in one business unit before going company-wide. The golden rule is identical for an SME: start with the employees who are already comfortable, measure actual usage (not intentions), formalise best practices, then extend. Being a week behind on one wave costs less than a month of failed adoption at full scale.

3. Integration into existing tools, no parallel stack

This is probably the most important lesson. None of these deployments creates an isolated "AI platform" — Claude slots into the dev tools (IDE, Git), the business tools (CRM, ERP), the communication flows (Slack, Teams). The end user sees their usual tool, augmented. For an SME, that means: before signing up for a new AI subscription, check that the solution integrates with your current stack (Microsoft 365, Google Workspace, your CRM, your PIM). An AI in a separate tab ends up in the cupboard. An AI inside the daily tool becomes invisible — and therefore adopted.

"The real lesson of the chaebols isn't their size — it's their discipline. And discipline doesn't depend on volume."

In the end, the three disciplines converge on a single principle: AI isn't deployed, it's framed. Framing by business function, framing by cohort, framing by integration. That's exactly what we structure month after month with our clients in the Codito AI Partnership — and it's also what you'll see in action in Part 3.

Book a strategy call

Discover the Codito AI Partnership

To close this edition, a fresh look behind the scenes — at a construction SME whose story loops straight back to what we've just covered on Samsung and NAVER.

On Monday 16 June, I led the first Codito BTP AI Partnership session at Gustave Rénovation, a high-end general contractor operating from three offices (Paris, Bordeaux, Vendée) — with, by design, lean teams of fewer than six people per office. An hour and a half with Amaury (head of the company) and Joseph (tech lead). A format deliberately different from what I've just described: no arbitrating between ten possible use cases. Just listening.

Three things this first session delivered

• A vision that was already clear. Amaury had prepared a homemade HTML prototype of his dream tool, and a specification born of several months of thinking. The decision-making discipline we credited to the chaebols in Part 2 — it already existed at Gustave before we arrived. The session wasn't there to sort — it was there to qualify.
• A stack to respect. Kalitix (quoting and invoicing), Excel, SharePoint, Outlook. The classic trap would have been to propose a separate AI platform. The conviction we shared is the one from Part 2: no parallel stack. What we build must slot into the existing tools — and progressively absorb the components that no longer deliver enough value.
• Non-negotiable technical sovereignty. Amaury requires that the solution be able to switch between Claude and Mistral depending on the constraints (GDPR, model quality, cost). That's exactly what our clients are starting to demand — and it's what separates an AI project designed for the next 12 months from one designed for next quarter.

By the end of the session: an AI maturity score measured across the five dimensions (presented in a previous edition), an NDA signed within the following 24 hours, a documentation stack already flowing in, and an M2 locked in for 5 July to present the first prototype. No vague promises. A framework.

The trap we sidestepped: promising technology At M1, the temptation to show what you can do is strong — demos, prompts, prototypes. We deliberately resisted it. The internal rule of the Codito BTP AI Partnership: the first session is devoted to understanding, not selling. The prototype will come at M2. And it will come by drawing on the most precious material we collected this week: the fine-grained knowledge of a trade — that of the high-end general contractor — which isn't on the internet, and never will be.