π

Intralinks · Latam Sales · May 2026

Getting real
with AI.

Trends · Fluency · Practical Usage

Vinicius Galera · Partner & Global VP of AI

scroll
01 / 38
π

Nice to meet you,

I'm Vinicius
Galera.

Global VP of AI & GTM

Tech enthusiast

01

Management Consultancy

Strategy, transformation and operating-model work across enterprise clients.

02

Technology Scale-up

Leading the AI GTM motion at a company that scaled from 900 to 9K people globally.

03

Continuous Study

Oxford University · Imperial College · Hebrew University of Jerusalem.

04

AI Community

Global Ambassador for Lovable and Anthropic · member of the Oxford AI Society.

02 / 38
π

The next 30–40 minutes

Two things,
nothing more.

01

What's happening in AI.

A grounded look at where the field actually is — beyond the hype cycle.

02

How I'm using it.

Real workflows, real tools, real outcomes — from my day-to-day.

03 / 38
Part I

What's happening
in AI.

The platform shift, the data, and why this moment is different.

04 / 38
π

AI ≠ ChatGPT, Claude, Lovable, Gemini…

"AI is whatever machines
can't do yet."

— Larry Tesler, 1970

Then, it becomes known as software.

AI is different than the chat apps you use — much bigger, and much older than 2022.

05 / 38
π

Andrew Ng · Stanford GSB · July 2023

AI is way bigger
than ChatGPT.

What everyone calls "AI" today — LLMs, agents, GenAI — is just one small, nascent slice of a much larger field that has been evolving for 70+ years.

Machine Learning dominates the surface. Generative AI is still the new arrival. Agentic AI is younger still — barely out of the lab.

Artificial Intelligence — the entire field
Machine Learning — supervised, unsupervised, reinforcement
Generative AI — LLMs, transformers, SLMs
Agentic AI — nascent, emerging today
Artificial Intelligence ML MACHINE LEARNING GenAI LLMs · SLMs Agentic NASCENT Reinforcement LEARNING YOU ARE HERE
06 / 38
π

Mental model 02 — the shift that changes everything

LLMs don't
know.
They predict.

Every word an LLM outputs is a probability distribution over possible next tokens. It's not retrieving facts — it's generating the most statistically likely continuation of your input.

This is why it can be brilliant and wrong simultaneously. It's also why how you prompt changes everything — you're shifting the probability landscape.

🎲Same prompt, different answers. Stochastic by design. Set temperature to 0 for determinism — but creativity disappears.
🔬Confidence ≠ correctness. The model states falsehoods with the same fluency it states truths. This is hallucination.
🌊Agents compound uncertainty. Each step multiplies the error rate. Human checkpoints are architecture, not caution.
Use freely
Drafts · Ideation · Research · Summarization · Options
✓ probabilistic fine
Use with oversight
Contracts · Analysis · Code · Client comms · Pricing
⚡ verify before sending
Never rely alone
Legal facts · Medical · Financial calculations · Compliance
✗ stakes too high
Agent chain — why human checkpoints are architecture
90%step 1
×
90%step 2
×
90%step 3
×
90%step 4
=
65%overall
Four steps at 90% accuracy each = 65% chance the final output is fully correct. This is why agents need checkpoints — not because AI is bad, but because probability compounds.
07 / 38
π

Why now · structural read

The hype is real AND inflated.

Four structural drivers — not opinions.

01
$500B+
Capital + Narrative Loop

Committed to chips, data centres, power. The capital needs the story — and the labs supply it.

02
7mo
Delegation Threshold

Task-length capacity doubles every 7 months (METR '25). 4 hours unsupervised ≠ 4 minutes.

03
1 = team
Rise of the HI-C

The High-Impact Individual Contributor. Zero reports, department-level output. AI as "average intelligence" — plus your craft — ships without the coordination loop. Elena Verna · Lovable '26

04
84% vs 27%
Social Proof Cascade

CEOs who say AI will change their business vs. those whose workforce is ready. The gap is anxiety.

Capital demands narrative. Capability crosses a threshold. Building gets cheap. Everyone feels behind. — Let's see the reality without the hype.

07 / 38
π

Driver 01 · the narrative loop

The pack converges.

Selected frontier LLMs by aggregate benchmark score, Jun 2022 – Jun 2026. OpenAI, Anthropic, Google, Meta and Chinese labs converging.
01

No moat. The race ends in commoditisation — DeepSeek matched GPT-4 for cents.

02

FOMO is the product. Model choice matters for ~5% of cases. Anxiety is the business model.

03

Systems compound. Models don't. Workflow + context + fluency — that's the moat.

08 / 38
π

Driver 02 · the delegation threshold · 1 / 2

The autonomy is real — and accelerating.

METR: time-horizon of software engineering tasks LLMs can complete 50% of the time. GPT-2 to GPT-5, near zero to over 2 hours.

0 → 2 hours of unsupervised work — in under three years.

GPT-2 found a fact on the web. GPT-5 exploits buffer-overflows. METR '25

09 / 38
π

Driver 02 · the delegation threshold · 2 / 2

The tech is already good. Build for stage 3. Design toward stage 4. Don't promise stage 5.

Our World in Data: AI test scores vs human performance across six capabilities, 1998–2023.

Already at or above human on 6 domains.

Reading. Image. Language. Handwriting. Speech. Reasoning closing in. OWID · Kiela et al. '23

10 / 38
π

Driver 03 · the rise of the HI-C

1=team

High-Impact Individual Contributor

“The real flex isn't the VP title anymore. It's the IC who ships what a whole team used to — with no direct reports, paid like a leader.”

— Elena Verna · VP Growth, Lovable · May '26

Everyone in this room can build now.

01 An idea is enough to start.
02 AI fills the average. You bring the craft.
03 No team, no roadmap meeting, no permission.
11 / 38
π

Reality check · the agentic org

The billion-dollar one-person company isn't here yet. We're moving toward it. Still too early.

McKinsey five-stage framework from automation accelerating tasks to AI agents replacing leadership. Estamos Aqui marker on stage 3.
01 Automation accelerates 02 Automation replaces tasks 03 Agents accelerate — estamos aqui 04 Agents replace supervision 05 Agents replace leadership

Build for 03. Design toward 04. Don't promise 05.

12 / 38
π

Everyone's tried it — adoption isn't there yet

Everyone's tried it.
But adoption is not fully there.

900M
Weekly ChatGPT users. 10% of all global adults. As of late 2025.
OpenAI / NBER · 2025
80%
Of users send fewer than 3 prompts per day. A mile wide. An inch deep. — Benedict Evans
OpenAI Wrapped · 2025
73%
Of ChatGPT usage is now non-work related. AI has escaped the office.
OpenAI / Harvard NBER · 2025
9.7%
Of US firms use AI in production. Up from 3.7% in 2023. The vast majority still aren't in.
US Census Bureau · Aug 2025
US workplace AI use by industry, June 2023 to December 2025. Tech leads at ~75% any-use, followed by Finance and Professional Services. Healthcare, Retail, Manufacturing and Government remain under 45%.
US workplace AI use by industry · June 2023 – December 2025 · Any / Weekly / Daily use
08 / 38
π

Reality check · deployment hasn't happened

AI hasn't been deployed yet. It's still a tool that only a few use. The headlines are loud — the rollout is quiet.

OpenAI, Anthropic and Google enterprise AI deployment ventures backed by TPG, Bain Capital, Brookfield, Advent, Blackstone, Hellman and Friedman, Goldman Sachs, KKR and EQT.

The labs are signing the deals. The capital is moving. The deployment hasn't happened.

14 / 38
Part II

How I'm using it(and maybe you could too).

From theory to practice — the workflows, prompts, and habits that actually work.

15 / 38
═══════════════════════════════════════════════════════════ -->
π

Anthropic · 81,000 users · 159 countries · Dec 2025

They don't want
faster email.

The largest qualitative AI study ever conducted. 70 languages. One question: "If you could wave a magic wand, what would AI do for you?"

The #1 answer was not productivity. It was time back for things that matter.

19%
Professional Excellence
Less busywork. More meaningful work. Not just being faster.
14%
Personal Transformation
Emotional growth. Mental health. Self-understanding.
11%
Time Freedom
"Last Tuesday, AI let me cook with my mother instead of finishing tasks."
67%
Net positive sentiment
But 27% fear hallucinations, 22% fear job loss. Hope and fear coexist.
09 / 38
π

My actual daily stack

How I run on AI
every single day.

🧠
Claude (Anthropic)
My primary thinking partner. Strategy decks, client frameworks, research synthesis, writing, devil's advocate. I don't open Google before opening Claude.
Daily
🛰️
Antigravity
Agentic dev environment. Spin up autonomous coding agents that plan, build and ship full features.
Daily
💻
Claude Code
Terminal-native coding agent. Refactors, builds prototypes and ships PRs straight from the CLI.
Daily
🎯
Lovable
Describe UI, get production React. I'm a Lovable ambassador.
Daily
🎙️
Granola
Every meeting auto-transcribed + summarized. I never take notes manually.
Every meeting
🔍
Clay + Perplexity
Prospect research, contact enrichment, competitive intelligence at scale.
GTM
📎
Paperclip
Always-on context capture. Clips, notes and references flow into one searchable AI memory.
Context
10 / 38
π

My goal

Build AI systems,
not just tools.

At scale — running in production, daily

5 AGENTS
🧭
AI Chief of Staff
Daily brief from email, calendar, WhatsApp + backlog — synthesised before I open anything. Includes a Presentation Maker: content in, deck out, no designer.
💼
AI Chief Revenue Officer
CRM consolidates every client touchpoint, enriched live with Clay + Apollo. Granola transcripts auto-pushed as structured briefs. A Personal Researcher writes outreach that sounds researched — because it is.
🧠
My AI Board
Cloning great leaders from their public sources. I ask the board a question and listen to perspectives from Buffett, Dalio, Naval, Amodei, Einstein, Curie, Freud and Jung.
⚙️
Automation Builder
The meta-agent. Describe the workflow you want — it builds it. You own the result without learning the tool.
🎓
My MBA
Research → NotebookLM → podcast → Spotify → listen on a walk. 4-week learning sprints, with the Chief of Staff generating a personalised checklist for each session.

Experimenting — the frontier

2 BETS
📎
Company Twin
Clone the company with AI agents. Every decision, meeting, document captured and queryable. The org's intelligence survives turnover.
🧬
Vini's 2nd Brain
Personal knowledge graph. Everything I've read, thought, decided — connected and searchable. Obsidian-style, AI-native.
17b / 38
π

At scale — 1 / 7

🧭 AI Chief
of Staff.

The operating layer of my day. It decides what reaches me, in what order, and in what shape.

Daily brief, before I open anything
Email, calendar, WhatsApp + backlog — synthesised into one read.
Presentation Maker
Content + design system in — deck out, in real time. No designer dependency.
I start the day focused, not triaging
~60–90 min of inbox + planning work absorbed by the agent.
Daily briefing — calendar + inbox synthesis
Inbox FYI + backlog status
17c / 38
π

At scale — 2 / 7

💼 AI Chief
Revenue Officer.

The full GTM intelligence stack. Every touchpoint with a client or prospect, unified.

AI CRM
Conversations, emails, meetings, notes — enriched live with Clay + Apollo.
Meeting Intelligence
Granola transcript → structured brief → auto-pushed into the CRM.
Personal Researcher
Outreach that sounds researched — because it actually is.
AI CRO account view
AI CRO pipeline report
17d / 38
π

At scale — 3 / 7

🧠 My AI
Board.

A personal advisory board I can convene on demand. I bring a real decision; eight minds challenge it from eight different angles before I move.

Different lenses, same problem
Ray for diagnosis, Naval for the games being played, Freud for what I'm not saying, Marie for the work itself.
It challenges the framing first
Before answers, it questions whether I'm even working on the right problem.
Round two when I'm ready
Warren, Dario, Jung, Einstein — sharper once the decision is properly framed.
Ray Dalio Naval Freud Marie Curie Warren Buffett Dario Amodei Jung Einstein
AI Board convening on a real decision
17e / 38
π

At scale — 5 / 7

🎓 My MBA.

Continuous learning, systematised. The aspiration most people never execute — running on autopilot.

Research → podcast → walk
Content into NotebookLM → podcast generated → published to Spotify → I listen while moving.
4-week learning sprints
Chief of Staff generates a personalised checklist per session, based on current context + priorities.
Learning compounds without scheduling it
Dead time becomes deep time. Every week I'm sharper on what matters now.
NotebookLM
Spotify episode
▶ Listen on Spotify — AI, explAIned · Digital Transformation & AI in Business
17g / 38
π

Experimenting — 6 / 7

📎 Company
Twin.

Clone the company with AI agents. The institutional-memory layer most orgs only realise they need after someone leaves.

Every decision, queryable
Meetings, documents, choices — captured and searchable in plain language.
Org intelligence survives turnover
Context doesn't walk out the door with a person.
Onboarding goes from months to days
New hires ask the org, not 12 people on Slack.
17h / 38
π

Experimenting — 7 / 7

🧬 Vini's
2nd Brain.

Personal knowledge graph. Everything I've read, thought, decided — connected and searchable. Obsidian-style, AI-native.

My memory becomes infinite + searchable
Nothing I've ever read is ever fully lost.
Connections I would have missed
The graph surfaces links between ideas across years and domains.
Thinking partner with my own context
An AI that knows what I already know — and pushes from there.
17i / 38
Interlude

So, in summary…

Pulling the threads together — what all of this actually means.

25 / 38
π

Recommendation 01 · Where to start

Focus on
AI fluency.

82% of HR leaders now prioritise AI literacy. 65% of employees don't feel confident. The gap isn't capability — it's fluency.

DelegateKnow what to hand off — and what to keep.
DescribeGive AI the context it needs to succeed.
DiscernEvaluate output critically — never accept blindly.
DiligentVerify, iterate, improve — treat prompts like code.

Anthropic · AI Fluency Framework — free course

Anthropic AI Fluency course QR code
anthropic.skilljar.com
AI Fluency · Framework Foundations
Take the course →
27 / 30
π

Recommendation 02 · The mindset shift

Stop using LLMs
like Google.

One-line prompts. Single answers. Restart every conversation. That's search behaviour. LLMs reward the opposite — depth, iteration, and context.

01
Learn Prompt Engineering
Role, context, task, constraints, format. The five ingredients of a senior-level prompt. Treat it like a craft, not a search bar.
02 · do this first
Reverse Prompt First
Before asking AI for an answer — ask it what it needs to know. "What are the 5 questions you'd need answered to make this exceptional?"
03
Build Skills
Document what only you know — judgement, frameworks, patterns. Turn tacit expertise into reusable prompts. Your skills become assets that scale.
04 · the unlock
Manage Context
Context is king. The same model gives a junior answer or a senior answer depending on what you put in front of it. Curate it like infrastructure.
28 / 30
π

Recommendation 03 · How to actually build & automate — by area, not by org

The AI Maturity Map.
Work backwards from the target level.

Don't climb the ladder one rung at a time. Pick the target level for each area, then assemble that level's building blocks from Day 0. The infrastructure that sustains N4 doesn't emerge from optimizing N1 or N2 — it has to be designed in.

N1
Individual productivity
Each person uses AI to optimise their own workflow. Gains concentrated in outliers.
Trap: "more tokens = more AI-native"
N2
Team productivity
Shared agents & skills inside workflows still designed for humans. The outlier gain becomes the team standard.
Trap: "more agents = more AI-native"
N3
Contextualised OS
A single agentic layer with the area's context. One person can operate the work of an entire team.
Trap: treating every context source the same
N4 · target
Decision intelligence
The layer proposes decisions; humans approve, edit, reject — and that signal trains the system. Quality converges to the best judgment in the area.
Trap: weighting all feedback equally
N5
Adaptive intelligence
The layer learns autonomously from outcomes. Results improve month after month without specific human intervention.
Not yet observed at scale
Principle 01
Don't climb. Work backwards. Pick the target level first — then assemble its building blocks from Day 0.
Principle 02
Think by area, not by org. Different areas live at different levels at the same time — that's the natural state.
Principle 03
Close the loop from Day 0. Open loops scale execution. Closed loops scale quality — and quality is what compounds.

Source: comp.vc — AI Maturity Map · comp.vc/landing-pages/ai-maturity-map

29 / 30
π

Thank you

Thank you.

Just start building and learning — everyone is on the same journey.

Vinicius Galera · Partner & Global VP of AI
Let's connect
linkedin.com/in/vinicius-galera

LinkedIn QR Scan · LinkedIn
30 / 30