LOCCO

1Framing: where I agree and where I push back

Your hypothesis, stated fairly: AI should be structural rather than peripheral; humans should concentrate on vision, brand, creativity, negotiation, and capital, while AI and automation progressively absorb analytical, operational, coordination, and documentation work, producing a company that is lighter, faster, more scalable, and more capital-efficient than a traditional one.

[Assumption] I read LOCCO as the new international house being built by the team behind David Locco: jewelry and fragrance on a shared brand of craft and personal expression. If the relationship or the model is different, tell me and I will re-tune the brand-risk sections; the operating logic below stands either way.

Where I fully agree:

• Day-zero beats retrofit. This is genuinely your edge. Most companies struggle because they bolt AI onto fixed processes and org charts; you can design the process around the capability. There is real economic upside on the table: AI-first operating models are projected to deliver materially higher output per person than traditional ones (WEF).
• Your human boundary is correct. Vision, brand, creative direction, negotiation, and capital allocation should stay human. That is exactly where judgment, taste, and relationships compound, and exactly what AI is worst at.

Where I push back:

1. “AI-native” as written sounds like a tooling claim. It isn’t. The companies that win don’t have better models than everyone else; they have better process and data discipline, so the same models actually produce value. I will keep returning to this: AI-native is an operating discipline, not a tech stack.
2. “AI absorbs the work” understates where the difficulty moves. AI can draft, analyze, and coordinate, but judgment, accountability, and quality control don’t disappear. They concentrate onto fewer people who must now review more. If you don’t design for that, your senior people become bottlenecked reviewers and the promised speed evaporates.
3. “Lighter, faster, cheaper” has a failure mode: “thin, fragile, undifferentiated.” Optimizing to look AI-native can actively harm a craft brand. The goal is leverage, not minimalism for its own sake.

Holding the ambition and these corrections together is the whole game.

2A testable definition of “AI-native”

Slogans don’t survive contact with a busy week. Here is a definition the team can actually hold:

Four concrete signals tell you whether it’s true:

These four signals become the backbone of how we measure progress (§9).

3Critical assessment of the hypothesis

What is right is covered above: the day-zero advantage and the correct human boundary. Here is the harder part. What is naive or under-specified:

• The bottleneck is misdiagnosed. The briefing implicitly treats access to AI as the lever. The evidence is blunt: failure is organizational and data-driven, not technical (MIT; Qlik). Giving everyone Claude without deciding where knowledge lives is like hiring brilliant analysts and giving them no files to work from.
• AI-as-identity vs. AI-as-leverage. There is a temptation to make “we are AI-native” part of the brand story. For a luxury house, be careful: customers in jewelry and fragrance are paying partly for the human. The industry consensus is that AI works as augmentation and backfires when it visibly replaces craft. Be AI-native in how you operate; be human in how you show up.
• The “absorb coordination and documentation” promise hides the hardest part. Documentation and coordination are where most companies’ tacit knowledge lives. Automating them well requires that the knowledge first be made explicit, which is real, deliberate work, not a switch you flip by adopting a tool.
• No definition of “better.” The hypothesis lists virtues (lighter, faster, more disciplined) without a way to know if they’re being achieved. Without that, you can’t tell leverage from theater. §9 fixes this.

This is a strong hypothesis that needs a sharper diagnosis of the bottleneck and an explicit guard around the brand. Both are fixable now, cheaply, because you’re early.

4The recommended operating model

Think in three layers, with one clear rule for routing work between them.

The routing rule: new or ambiguous work starts human-led; once a pattern is proven, AI assists it; once it’s truly mechanical and high-volume, it’s automated. Never automate work you don’t yet understand; that is how you encode mistakes at scale.

The single highest-leverage thing LOCCO can do this quarter is not “use more AI.” It is to decide where knowledge lives and start writing decisions down in a queryable form. A useful mental model: AI is a brilliant new hire who has read everything public but knows nothing about LOCCO specifically. Its usefulness is capped by how well you’ve written down how LOCCO actually works.

5Human-led vs. AI-assisted vs. automated: the function map

The boundary should be a principle, not just a list, so it extends to functions you haven’t created yet:

[Assumption] This map assumes a fairly digital, DTC-leaning operation with contract manufacturing. If you’re more retail- or own-production-heavy, the human-led set grows and the automation set shrinks; flag it and we’ll redraw the map together.

The headline: automate the back of the house aggressively; keep the front of the house human.

6Redesign from scratch vs. optimize

Redesign from scratch (because you’re early and can):

• Knowledge management: one queryable home for everything, designed for AI to read, from day one.
• Decision logging: a lightweight habit of writing down decisions and their reasoning. The highest-ROI new process you can build.
• Reporting & dashboards: generated and summarized by AI rather than hand-assembled.
• Onboarding: new people learn by querying the company’s written brain, not by shoulder-tapping.
• Customer service architecture: designed as AI-tier-1 + human-escalation from the start.

Optimize, don’t rebuild: creative, design, and negotiation workflows: augment with AI inputs, but keep the human core intact. Rebuilding these “for AI” is where craft brands lose their soul.

7Team, governance, and tool stack: the early stage

Team shape. Small, senior, and high-agency. The AI-native founding pattern is well established: at minimum, one person who deeply owns customer/market/brand, and one who is genuinely comfortable with data, automation, and AI tooling. Hire full-stack operators, people who own a whole domain end-to-end with AI as leverage, rather than narrow specialists who hand work across silos. The hiring bar shifts from “does this person have function-X experience” to “does this person have judgment, and adapt fast.”

Governance: the part almost everyone skips, and where the 95% die. You need answers, written down, to:

• Where is AI allowed to act unreviewed, and where must a human approve? Default in stage 1: a human approves anything customer-facing or irreversible.
• Who owns AI output quality? Accountability cannot be “the AI.” A named human owns every AI-assisted output that ships.
• Data, IP, and confidentiality rules. Designs, supplier terms, and fragrance formulas are high-stakes IP. Decide what may and may not go into which tools, written down before, not after, an incident.
• A “decisions are logged” norm, modeled by leadership first.

This doesn’t need to be heavy. One page of clear rules beats a policy binder nobody reads. But it must exist before broad rollout.

Tool stack: deliberately minimal. Tool sprawl is a documented failure mode. For stage 1: Claude as the horizontal layer the whole team uses for thinking, drafting, research, and analysis; one home for knowledge (a single docs/wiki system); and one connective automation tool when you’re ready to wire flows together, not before. Resist everything else until a real, repeated need proves itself. Add tools to remove a felt pain, never to feel modern.

8The phased roadmap

The sequencing principle governs everything: adoption → patterns → automation. Never automate before you understand the work.

Most companies try to start at Phase 3. Starting at Phase 0 and earning your way down is the whole advantage of being early.

9How we’ll measure success

The purpose of measuring is to avoid the trap I warn about in §12, efficiency theater: celebrating activity while outcomes don’t move. Tie metrics to your own stated goals:

Vanity metrics to explicitly ignore: number of AI tools deployed, number of prompts sent, “% of team using AI.” These measure motion, not progress.

Caveat: in Phase 0–1 these are directional signals, not KPIs. Measuring too hard, too early kills the experimentation you need first. Track them lightly until adoption is real, then tighten.

10The role of Claude in initial adoption

Concrete, not slogans. In the first stage, Claude should be:

• A research and synthesis engine: market scans, competitor analysis, regulatory and material research, distilled to what matters.
• A drafting partner: first drafts of copy, specs, briefs, proposals, internal docs. A human edits for brand voice and judgment.
• An analyst over your own information: reasoning across your data, documents, and decisions. This is why the foundation matters: Claude is far more valuable reading LOCCO’s context than generic web knowledge.
• A documentation engine: turning meetings and decisions into searchable records, which directly feeds the knowledge foundation.
• A thinking partner: a place to pressure-test strategy, stress an argument, or generate options before a human decides.

Practically, this means a Claude Team workspace with shared Projects per function, each preloaded with LOCCO’s brand voice, decision log, and SOPs, so Claude answers as LOCCO, not as the generic internet. Setting this up well takes a day and multiplies the value of everything above.

What Claude should not be yet: an unsupervised actor taking real-world actions without review; the system of record (it complements your knowledge home; it isn’t it); or a substitute for hiring senior judgment: it amplifies good judgment and amplifies bad judgment just as fast.

The adoption mechanic that matters most: the weekly “well / badly” loop. It converts scattered individual experiments into shared, compounding organizational skill, and it surfaces, from real use, exactly which processes are ready to codify and automate.

11What leadership and the team must change

Leadership:

• Model the behavior. AI-native fails the moment leaders exempt themselves. If you want decisions logged and work drafted with AI, you do it first, and visibly.
• Accept more explicitness, not less. AI-native means writing down more of what used to live in your head. That is a real change of habit, and it has to start at the top.
• Protect the foundation work. It’s unglamorous and easy to deprioritize against shipping. Don’t. It is the thing that makes everything later possible.

Team: the new baseline skill is working with AI, a core competency, like email once was. The hiring bar shifts toward judgment and adaptability over narrow function experience.

The cultural risk to manage actively (see §12): fear. If people believe AI is here to replace them, they will quietly not adopt it, and the whole thing stalls. The framing has to be true and repeated: AI is leverage that makes each person bigger, not a headcount-reduction program. In a small team, morale is the rollout.

12Risks, limits, and blind spots

This is the most important section for credibility, so I’ll be direct.

1. Brand erosion: the big one for LOCCO. Over-automating the front of the house erodes the human authenticity luxury customers pay for, and in your case the promise is explicit: handcrafted is in the house’s own tagline. That promise is exactly what careless automation would quietly spend. Industry consensus and the data both say AI works as augmentation and backfires as a visible replacement for craft. Mitigation: automate the back, keep the front human; never let AI be the last touch before a customer.
2. Over-automation and fragility. Automated processes nobody understands break silently and take tacit knowledge with them. Mitigation: the “never automate what you don’t understand” rule, and humans who can still do the work manually.
3. Quality and hallucination accountability. AI is confidently wrong sometimes. Mitigation: a named human owns every shipped output; human-approval gates on anything customer-facing or irreversible.
4. Data, IP, and confidentiality. Designs, supplier terms, and fragrance formulas are high-value IP. Mitigation: explicit rules on what goes into which tools, written before rollout.
5. Efficiency theater. Measuring AI activity instead of business outcomes. Mitigation: the §9 metrics, and the named vanity-metric blocklist.
6. Hiring for tools, not judgment. AI raises, not lowers, the premium on good judgment. Mitigation: hire for judgment and adaptability; keep the team small and senior.
7. People risk: the silent killer. Team fear of replacement → quiet non-adoption, attrition, morale collapse. In a small team this single-handedly kills an AI-native rollout. Mitigation: the leverage framing, leaders adopt first, and being good at this is made visibly career-enhancing.

The limits of this very advice: where my recommendation breaks down

• If LOCCO is far more retail- or own-manufacturing-heavy than I’ve assumed, the human/automation balance shifts materially toward human, and the back-office automation upside shrinks.
• If the early team is non-technical and can’t self-serve tooling, Phase 0–1 needs more hands-on enablement and a slower ramp than implied here.
• If the launch timeline is very tight, you may not have room to build foundations first, in which case we consciously take on foundation debt and plan to repay it, rather than pretending we did the groundwork.

Naming where I might be wrong is deliberate. An advisor who can’t tell you the limits of their own advice is selling you something.

13Closing recommendation

Be ambitious in vision and disciplined in sequence. The ambition, an AI-native LOCCO that is lighter, faster, and more scalable than a traditional house, is the right target, and being early is a genuine edge. But the path that actually gets you there is unglamorous:

Do that, measure outcomes rather than activity, and protect your team’s belief that AI makes them bigger, and you’ll have something most “AI-native” companies only claim to be: a company structurally built for speed, control, and international scale, with AI in the operating system rather than bolted on the side.

Part II turns this into a week-by-week plan. Part III describes how we could do it together.

◆Seven open questions

The answers to these move this from a strong general framework to LOCCO’s specific operating plan; they are the agenda for our first working session:

1. Stage & traction: pre-launch, launched, or generating revenue? Timeline to (next) launch?
2. Team today: how many people, covering which functions, at what technical comfort level?
3. Product & production: own production or contract manufacturers? How much of the value chain is in-house vs. outsourced?
4. Channel: DTC, retail/wholesale, or both? Which markets at launch?
5. Existing systems & data: anything already in place, or genuinely greenfield?
6. Budget posture: appetite for tooling and automation spend in stage 1?
7. Brand stance on AI: visible in the brand story, or strictly an internal operating capability? This materially changes the front-of-house decisions in §5 and §12.

II.1The first 90 days, week by week

Day 1, literally

1. Create the Claude Team workspace and invite everyone.
2. Send the team a three-line “why we’re doing this” note (template in II.4); the framing matters more than the tool.
3. Book the 90-minute kickoff for this week.
4. Create one page called Decision Log in whatever tool you already have.
5. Log decision #1: “We are building LOCCO AI-native; here is why.”

Weeks 1–2 · Turn it on (no automation yet)

Goal: everyone uses Claude daily on real work, and we decide where knowledge lives.

Don’t yet: build automations, buy extra tools, or write policies longer than one page.

Weeks 3–4 · Build the habit, capture what works

Weeks 5–8 · Codify the wins

Weeks 9–12 · First careful automation

II.2Role playbooks: what each person actually does with Claude

For each role: concrete use cases, plus a real example prompt to paste and adapt.

II.3Starter prompt library

Keep these in the knowledge base; everyone can use and improve them.

Prompt tip for the team: good prompts = context + role + task + format. Tell Claude who LOCCO is, what role to play, exactly what you want, and what shape the output should take.

II.4Templates

The Decision Log: the highest-ROI habit

One row per real decision, kept in the knowledge base:

The weekly “AI wins & misses”: 30 minutes, every week

1. Each person shares one win: a task AI made faster or better.
2. Each person shares one miss: where it failed or wasted time.
3. Wins → the prompt goes into the library. Misses → note what not to use it for.
4. Anything repeated 3+ times → automation candidate.

The one-page governance rules

Team announcement note: Day 1, three lines, from the founder

Knowledge base starter structure

LOCCO Knowledge Base
├─ Decision Log
├─ Prompt Library
├─ Role Playbooks
├─ SOPs & Processes
├─ Brand: voice, guidelines, assets
└─ Reports & Metrics

II.5The stage-1 tool stack: named, minimal

Resist tool sprawl. This is enough to start:

Add nothing else until a real, repeated pain proves it’s needed. Every new tool is a tax on a small team.

• We operate this way ourselves. ORB Global runs AI-native internally; the advice is practiced, not theoretical.
• We build. Most AI advisors deliver decks. We deliver working systems, and we’ve done the unglamorous parts (data, process, automation) for years.
• We told you where we’d be wrong. Part I names its own assumptions and limits. That is the standard we’d hold through the whole engagement.

Commercials, deliberately simple: we’ll shape the scope together after the working session. No open-ended retainers, and no dependency by design: everything we install, your team can run without us.

Nima Golsharifi · ORB Global Ltd