Transform eToro to operate at the speed of AI.
Not as a goal — as a decision we make in this room.
Target 100x
Before we dive in — everyone in this room needs to be on the same page. This page is your single source of truth for the offsite: context, fleet links, participants, and the agenda.
Big Goal: Transform all of eToro to work at a speed of 100x. Not as an aspiration — as an operating reality, embedded in how we build, decide, and ship every day.
It is a new operating system for organizations. The companies that win won't be the ones that adopt AI incrementally — they'll be the ones that fully rewire themselves around it.
"Our biggest challenge has always been: how do we deliver everything our customers want under our limited resources? How do we plan resources and priorities? This is changing now — but only if we build the right infrastructure."
— Yoni AssiaSpeed is not a byproduct of transformation — it is the competitive advantage. The goal is to make speed itself the moat.
From months to daysThe old model layers AI on top of existing processes. We are refactoring the entire operating model so AI is the default, not the exception.
Embedded everywhereManual requirements → fragmented tools → trapped knowledge → slow iteration. This was the only way. Now there's a better way.
Exponential modelBenchmarks from Robinhood, Ramp, Klarna, Plus500 — AI-first operating models, 2025–2026 research
We're not starting from zero. The fleet is live and in production today. It learns, routes, executes, and compounds knowledge across every session.
Routes work across the full fleet of specialist agents. DevOps, Dev, Trade, Apps, QA, Comms, People — one entry point, infinite specialization.
Fleet in productionPersistent, cross-agent memory system. Every agent reads and writes to a shared knowledge base. Fleet skills hub at fleet.clawz.org/skills — daily learning loops compound.
v0.9.1 · 13 toolsInternal app store for AI-built tools. Product leadership already ships here. The hackathon apps live here.
builders.etoro.comSuperApp / Banking · Visa Benefits (LIVE) · CoinsClaw · ClawMM · AlphaEar · ProInvestorClaw · Splinter Segmentation AI · App Ventures Store
All in flightEvery decision, task, and workstream sits in exactly one layer. AI sits at the center — serving all four. This is not a hierarchy. It is an operating system.
Sets strategy and direction. Owns AI accountability and EU AI Act compliance. Approves transformation phases and org targets. Final risk sign-off. The smallest, most powerful layer.
Live dashboard — CEO sees one screen every week. No status meetings. Routing, coordination, first-draft outputs. Not decisions.
Designs the systems AI operates in. Sets rules and quality standards for agents. Builds the CI/CD harness and testing infrastructure. Critiques AI output — finds failure modes. Defines what "done" looks like for Builders.
Claude drafts comms, analyses data, writes phase reports. Cursor + Copilot in the harness build. Architects review — they don't generate from scratch.
Ships with AI. The primary output engine. AI agents are co-creator — not just assistant. Writes prompts, not code from scratch. Ships features and tests same day. Raises output 1.5×–2× without extra headcount. Reports exceptions up — no approval queue.
Every PR goes through AI-assisted review before a human sees it. Feature cycles measured in hours. AI is the primary builder; the human is the director.
Reviews AI output for risk and correctness. Investigates exceptions flagged by AI triage. Applies regulatory judgment (FCA / EU AI Act / MiFID II). Signs off on high-stakes decisions — human always in the loop. Escalates unresolved risk to Governance.
CS agent handles first-pass triage. Compliance agent flags regulatory issues before humans review. Finance sees live budget dashboard. AI generates; Operators approve.
Same model in every department — Product, Engineering, Finance, Risk & Compliance, Customer Ops, HR, Marketing. The model doesn't change by function. The tools do.
We can refactor at 50x. The bottleneck is never the code anymore — it's the infrastructure around the code. We need to ask: what do we need to build so that AI agents can actually work within our systems?
CI/CD with AI in the loop. Every PR reviewed by AI before a human sees it. Infrastructure agents can actually operate within — not just around.
Harness before exitsNative memory vs MemClaw — we need one unified memory layer. MemClaw ↔ Caura integration is on the agenda. Phase 1 rollout plan to be finalized this offsite.
Mon 14:00 sessionMulti-cloud and vendor agnostic moving forward. We need AI to be able to help us migrate and move between vendors easily and use always best in class. No lock-in — architecture that lets us shift workloads across providers seamlessly.
Ella · Day 2 10:00The "Refactor Faster" session is a mandatory discussion. We cannot unlock 100x velocity without Agent-Ready Systems. This is the infrastructure conversation that unlocks everything else.
— Yoni Assia, pre-offsite notesThe transformation happens at three levels simultaneously. Each layer amplifies the others — individual productivity compounds into team velocity, which compounds into company-wide intelligence.
Every employee has a personal AI agent. Automated workflows. Instant knowledge access. Less time on repetitive work — more time on high-impact thinking.
Tool: Claw / Claude / Cursor
Teams of One. Teams of Three. Small, autonomous units replacing large, slow departments. AI copilots for planning and execution. Faster delivery with fewer dependencies.
Tool: Fleet orchestration
Unified data layer. Cross-functional automation. Continuous learning loops. Better decisions, faster alignment, scalable growth without adding complexity.
Tool: MemClaw + Base44
One AI Champion per Business Unit. They run the 30-day challenge in their BU and have already passed it themselves. They coach others, flag blockers to their HRBP, and act as the living proof that the transformation works.
The 30-day challenge is not a test we administer. AI Champions in each BU have already run it themselves before asking anyone else to take it.
Move from "waiting for resources" to "App Ventures" and "Product Hackathon" mindsets. Standardize what works. Scale aggressively. Empower builders across every department.
Every employee can ship an AI-built product. The App Store is the distribution channel. The hackathon apps are the proof of concept. Liquidation Map is already live. Apps already in the pipeline.
clawz.org/appstoreDay 1 showcase: noteworthy apps from the AppStore Hackathon, created by product leadership and finalist teams. This is what "Teams of One" looks like in practice.
Day 1 · 13:15 sessionAt this offsite, every manager gets access to Claw. Every manager leaving this room with an AI fleet at their fingertips. This is how the transformation becomes self-sustaining.
Offsite revealBuilders define what "done" looks like before they start. Kill features that don't move metrics. AI usage rate is part of the output metric. Process, meetings, and decks are not outputs.
New operating standardThe question of ownership is the question of what this offsite is actually about. Who owns each project? What support do they need? How do we create the structure around the individuals who are already building? The goal is individuals — Teams of One or Teams of Three. Right now it still takes 20 people in many cases. That's the gap we're closing.
— Yoni Assia, pre-offsite notesKeep to the schedule. Move fast. Make decisions. Focus on the Art of Possibility — not the challenges.
This offsite has a mandate: move fast, make decisions. Each item below has an owner and a deadline. We leave with these closed.
| Category | Decision / Action | Owner | Priority | Deadline |
|---|---|---|---|---|
| Infrastructure | Define multi-cloud strategy: vendor-agnostic architecture, workload portability, AI-driven migration tooling | Ella | ● Critical | Day 2 10:00 |
| Infrastructure | Standards, governance, and deployment policy — formalized and documented | Yoni + Ella | ● Critical | Day 2 close |
| Memory | Sign shared memory architecture doc (MemClaw ↔ Caura schema + auth + boundary) | Clawz + Caura | ● Critical | Mon 14:00 outcome |
| Memory | Prepare MemClaw Caddy auth + IP allowlist options (A/B/C) from audit | MemClaw / DevOps | ● High | Mon 14:00 session |
| Memory | Resolve Caddy auth conflict (X-API-Key vs basic_auth) and deploy chosen model | MemClaw owner | ● High | Post-offsite |
| Memory | Expand MemClaw IP allowlist to cover full fleet access | Security + DevOps | ● High | Post-offsite |
| Memory | Technical requirements: Caura shared architecture, two-brain problem resolution | Yoav + Caura | ● Critical | Day 1 · 16:45 |
| Governance | Org flattening: draft which roles become agents vs stay human | Yoni + Solutions | ● High | End of offsite |
| Governance | Fleet security governance: confirm single firewall owner; document formally | SecurityClaw | ● High | Post-offsite |
| Governance | Are agents here to stay? Decision + rationale documented | All | ● Critical | Day 1 session |
| Training | Training plan for different tiers of AI expertise across eToro | Yoni + Solutions | ● High | End of offsite |
Not a dream. Not a roadmap. Not a slide deck.
The question is not whether AI will transform eToro.
The question is whether we lead that transformation — or catch up to it.
"A feature cycle that took six weeks now takes one day. We averaged 3–8 production deployments per day over 14 days. Bad features die the same day they ship. New features go live the same day they're conceived."
AI-First Market Research, 2025–2026 · Robinhood, Ramp, Klarna, Plus500