S2G Team AI Briefing
● S2G Investments · Team AI Briefing · July 2026
“Both OpenAI and Anthropic see capabilities increasing rapidly… Both labs see nothing above us but air.”
— Andrew Curran · The AI Daily Brief, “How the 4 New Models Will Change How You Work”
This isn’t a talk about AI risk or AI strategy. It’s about what’s sitting on your desk right now — and how to get more out of it.
● Monthly Update and AI Practice
01
Capabilities, expert benchmarks, job market reality, and the grounded rule for eliminating hallucinations.
02
The “Human Sandwich” framework, human judgment in loop, and leveraging lab compute subsidies.
03
Voice dictation, delegating loops, power habits, making web artifacts, Claude in Slack & Gemini Notebook.
04
Configuring Claude & ChatGPT, S2G Connectors (MCPs), RSI loops, the H2 2026 roadmap, and the AI Practice Program.
05
The hierarchy is becoming a network — the Organizational Singularity, and the 15-step playbook to run it.
● Team Survey
We know everyone is growing at a different pace, in different roles and responsibilities — and we want to make sure everyone has the training they need to be successful. A short survey is coming at the end of this session: tell us where you want to go deeper.
Getting more out of the everyday chat tools you already have.
Help setting up daily updates — calendar, inbox, and priorities each morning.
Agent-powered work sessions beyond the chat window.
Partner with someone from the tech team to develop a bigger tool for your workflow.
The coding agents — for analysis and automation, not just for engineers.
Whatever you want to learn about AI — the survey has a blank for it.
● Who does what
We’re here to help. Please reach out.
01
The frontier moved. Most people didn’t.
● Act 1 · Where the models are
“We’ve essentially found a way to make sand think… Perhaps 10x the industrial revolution at 10x the speed.”
— Demis Hassabis, Google DeepMind
“Possibly by 2026 or 2027… AI systems will be best thought of as akin to an entirely new state populated by highly intelligent people — a ‘country of geniuses in a datacenter.’”
— Dario Amodei, Anthropic
● Act 1 · Where the models are
Product desk · Anthropic
65% of the Claude team’s code
…is now written by Claude itself.
Google I/O · May 2026
3.2 quadrillion tokens a month
…across Google’s AI surfaces — a 7× jump in one year. — Sundar Pichai
Business wire · May 2026
$1B → $47B
Anthropic’s annualized revenue, Dec 2024 → May 2026 — the fastest software scaling in history.
Consumer desk · March 2026
900M people a week
…now use ChatGPT — up from 400M a year earlier.
Enterprise desk · 2026
40% and rising
of OpenAI’s revenue is now enterprise — tracking to match consumer by year-end.
● Act 1 · Where the models are
“So far at least, I’m pretty sure AI has been net job creating. This was not what I expected.”
— Sam Altman, OpenAI
Unemployment for 20–24 year olds: effectively unchanged since the AI boom began.
— The AI Daily Brief, “AI Optimism vs AI Pessimism”
“The more we automate, the more expert human work there is to do.”
— Dan Shipper · “Why Agents Still Need Humans”
● Act 1 · Where the models are
Under 2%
top models’ error rate on grounded summarization benchmarks today — down from ~3–27% across models in 2023.
Open-ended trivia recall from memory remains unreliable. Ask it to freelance — then verify.
02
The human is the point, not the leftover.
● Act 2 · This is your moment
Everyone at every firm is getting the same frontier intelligence. What can’t be downloaded is the judgment, initiative, and taste you bring to it — that’s where the differentiation lives.
— Andrej Karpathy, “Agency > Intelligence”
● Act 2 · This is your moment
Define the task and define what good looks like.
Hours of work compressed into minutes.
Apply taste, trade-offs, and responsibility.
“Entirely automating everything is not the future we want… As AI systems become more capable, the human role becomes more important: setting direction, making trade-offs, applying judgment and bringing values, taste, care and responsibility to work.”
— OpenAI, in writing · “OpenAI Declares the Next Phase”
“Human capital does not become less valuable as token capital grows. It only becomes more valuable… Without human direction, you have compute running in circles.”
— Satya Nadella, Microsoft
● Act 2 · This is your moment
Efficiency is great — we need it, and we’ll capture every bit of it we can. But the bigger opportunity is doing things you’ve never been able to do before.
Faster memos, quicker research, less grunt work. Necessary, and we’re doing all of it.
Interrogate an entire data room. Build a working tool without an engineer. Watch every signal in a sector at once. Things that simply weren’t possible before — that’s where the real value is.
● Act 2 · This is your moment
Creating content is now the easy part. The hard part is making it ingestible — something another person can actually absorb, and act on.
● Act 2 · This is your moment
One thing these tools don’t do is shrink the to-do list. When exploring gets cheap, the backlog becomes infinite — so the discipline shifts from doing more to choosing well.
● Act 2 · This is your moment
Not everyone is cheering. Data centers — the physical footprint of the AI boom — have become a lightning rod for frustration over power costs, water, transparency, and concentration of wealth. Worth keeping in view, as investors and as neighbors.
142
protests in a single day — the first nationwide action against the AI infrastructure buildout (July 2026).
42
states where those protests took place — a rare issue uniting both parties.
~65
state-level data center restrictions enacted since the start of 2025.
— The Washington Post, “How data centers became a symbol of Americans’ rage” · July 18, 2026
● Act 2 · This is your moment
AI is now statecraft, and the race dynamics are genuinely uncertain.
“All in all, the export control is a failure… it gave them the spirit, the energy, and the government support to accelerate their development.”
— Jensen Huang, NVIDIA, on China chip controls
03
Tactics you can use now.
● Act 3 · Tips and tricks
AI lives or dies on the context you give it. No matter how fast you type, you almost certainly talk faster.
*S2G has approved Wispr Flow — and right now both ChatGPT and Claude have audio input capabilities. And ChatGPT has gotten really good.
● Act 3 · Tips and tricks · Then, everybody else
Every function is carving off the pieces of the job that agents can run — the research, the drafting, the reconciliation, the reporting.
Marketing
Agent-assisted
Research
Agent-assisted
Legal
Agent-assisted
Finance
Agent-assisted
Operations
Agent-assisted
● Act 3 · Tips and tricks
“I don’t prompt Claude anymore. I have loops that are running… My job is to write loops.”
— Boris Cherny, creator of Claude Code
● Act 3 · Tips and tricks
/goals● Act 3 · Tips and tricks
Markdown is plain text with light structure. A human can skim it in any editor; an agent can ground on it, update it, and diff it. No databases, no exports — just files. This is the format your working knowledge will increasingly live in.
Flashfood — Board Deck Q2 2026
Highlights
Asks
● Act 3 · Tips and tricks
Ask Claude for a page instead of a file — a report, a dashboard, a working mini-tool — and it hands you a link. That’s an Artifact. (You’re looking at one right now.)
Send the link once. When the story changes, update it — everyone sees the latest version, no re-sending attachments.
Toggles, filters, calculators, checklists — readers explore the data, not just read about it.
No account, no install — opens on any laptop or phone.
Dashboards instead of spreadsheets. Proposals with ROI toggles instead of static exhibits.
Nothing is public until you share it — you decide who gets the link, and when.
A teammate can take a copy of yours and adapt it — no starting from scratch.
● Act 3 · Tips and tricks · Live example
A working example of the play on the previous page — an investor page built and kept current the way you’d otherwise maintain a deck. Opens in a new tab.
Wilkinson Baking Company
Fully automated · AI-native bakery
A New Model For Bread
Real bread from a few simple ingredients — baked across a distributed network of automated bakeries that learn from each other.
rich
● Act 3 · Tips and tricks
“The third major redesign of LLM UI/UX… a persistent asynchronous entity with org-wide tools and context working alongside teams of humans.”
— Andrej Karpathy, on Claude in Slack
● Act 3 · Tips and tricks
A research assistant that answers only from sources you give it — with citations. If the answer isn’t in your sources, it says so. Formerly : notebooklm.google.com. One notebook per deal, sector, or project; up to 50 sources of nearly any type.
| Feature | What it does | When to use it |
|---|---|---|
| Chat with citations | Q&A grounded in your sources only | Diligence on a data room; digesting a 200-page report |
| Audio Overview | Two-host podcast made from your docs | Listen to a deal memo on the drive in |
| Mind Map | Clickable visual map of the material | Getting oriented in an unfamiliar sector fast |
| Reports / study tools | Briefing docs, FAQs, flashcards, quizzes | Prepping for a board meeting or LP call |
CIM + financials + call notes → “what are the stated churn assumptions? cite them.”
Articles + talks → Audio Overview for the commute.
Notes + the attendee’s writing → briefing doc in one click.
● Act 3 · Tips and tricks · The deeper move
The conversation has moved past “which model?” Building with AI is now a systems-design question: route the right task to the right model at the right price.
the heavy hitter · $$$
the economical · $
the custom job · $$
● Act 3 · Tips and tricks
Two-week “pods”: one AI-proficient person + one domain expert.
15 hrs → 30 min
Capital allocation process
2 days → 10 min
Financial pacing
04
Building AI into the way you work.
● Act 4 · Your setup at S2G
Click a step to check it off — progress is saved on this device.
● Act 4 · Your setup at S2G
RSI — recursive self-improvement — is the thing at the labs: “AI doing AI research will become the determining factor of the pace of progress.” (OpenAI) Here’s your personal version of the loop:
● Act 4 · Your setup at S2G
Not a wall, a spectrum. Be explicit about where each output sits: personal leverage on one end, shared formats on the other.
A morning rundown of your day ahead. A summary of which emails still need answers. Fast, in your voice, tuned to how you work — nobody else has to read it.
IC memos, financials, quarterly reporting. One consistent format, so anyone can pick them up, absorb them quickly, and compare across projects.
● Act 4 · Your setup at S2G
Technology & AI — 2026 Roadmap
Chris Marshall, Head of AI and Tech · May 2026
Vision
Build institutional intelligence that compounds across every meeting, deal, and portfolio interaction — spend 80% of time on judgment rather than administration.
The upcoming training follows the maturity path: each level builds on the one before it, and H2 has a milestone for each.
● Act 4 · Your setup at S2G
● Act 4 · Your setup at S2G
Not homework — real work. Each month, every person picks one real deliverable and runs it through AI.
A memo, a model, a screen, a brief — something already on your plate.
Use the month’s workflow or tool as the starting point, not the afterthought.
Show the result — time saved, quality gained, or why it didn’t work — at the next session.
● Act 4 · Your setup at S2G
Hey — here’s a rough sketch of what the stack looks like, just to be helpful.
05
The hierarchy is becoming a network.
● Act 5 · The future is here · The ExO 3.0 playbook, made runnable
For a century, companies were built as human pyramids because coordinating people was expensive. AI just made coordination nearly free — and every pyramid is quietly turning into a mesh. We’ve turned that idea into a 15-step playbook you can actually run.
▲ Org chart, 1926 · ● Intelligence network, 2026
Open the full playbook →AI Training Check-in
A pulse check on how the S2G organization is progressing with AI
What level of comfort do you have with AI tools these days? *
How often do you use AI today? *
● Team Survey
Two minutes, right now: tell us where you want to go deeper, so the training meets you where you are.
Getting more out of the everyday chat tools you already have.
Help setting up daily updates — calendar, inbox, and priorities each morning.
Agent-powered work sessions beyond the chat window.
Partner with someone from the tech team to develop a bigger tool for your workflow.
The coding agents — for analysis and automation, not just for engineers.
Whatever you want to learn about AI — the survey has a blank for it.
● Close
“Sorry to anyone who thought AI would mean we’d work less, at least for now. AI makes it easy to explore more than you did before, and so you start doing far more as a result.”
— Aaron Levie · “Why Agents Make Every Job a Startup”
A
Reference for later.
● Appendix