Replying to @BonesawMD
If you’re really interested in carving your name in history you should take a moment to ask why? Can your fear of oblivion be overcome by focusing on local impact?
106
Replying to @disco___cat @DrCameronMurray
💯. Prefect is an honour given to kids who know how to glaze teachers.
23
Brave piece. Murder rates of women by male partners or exes are horrifying, but men who are subjected to violence or coercive control by their partners are in a real bind. Social stigma and fear of counter claims stop them seeking help. open.spotify.com/episode/3ud…
17
Endorse
I came to the conclusion that most SaaS should become an MCP server.
we’re getting the economics of AI software backwards. A lot of AI startups are building their own chat interfaces, agents, and workflows, essentially reselling tokens from the big labs at low to negative gross margins.
What if SaaS companies did the opposite?
Take HR software. Instead of building an AI assistant, become the best MCP server for HR. Expose tools to:
- Benchmark salaries against proprietary market data
- Run compensation reviews and calculate pay equity
- Model salary increases against a department’s budget
- Recommend promotions based on company policies
- Update employee compensation in the HRIS
- Manage approvals, permissions, and audit trails
Let ChatGPT, Claude, or Copilot handle the conversation, reasoning, and orchestration. Let the customer pay for the tokens.
The SaaS company keeps the valuable stuff: proprietary data, integrations, business logic, permissions, compliance, and the system of record.
You still charge a SaaS subscription, but you’re no longer paying to generate every answer. You can ship faster, with fewer engineers, and potentially restore software-like gross margins.
The next generation of SaaS might not have much of a UI at all. Just an incredibly good MCP server.
21
Fuel efficiciency < a shorter route
This was a great piece, this line particular very inciteful:
"The best models allow you to reach quality with fewer iterative loops, and thus often more cheaply than models that cost less per token.
Cheaper models, however you prompt them, often fail at getting you to clear the quality bar and you end up starting over again manually, a huge waste. Minimizing iterations is far more valuable than trying to save on per-token costs.
A contrarian rule of thumb for 2026 has been that the more expensive a frontier model is per token, the higher its ROI."
8
For deterrence to work economically, the expected cost (fine × probability of detection and successful enforcement) must exceed the expected profit from the diversion. That’s a big number.
NEW: We are short Nvidia $NVDA. We believe Nvidia has a significant China problem. Please see our full report and disclosures now available on our website, link in bio.
54
Good list
I asked a PE operating partner what the biggest AI win across his portfolio was.
I thought he was joking. “Timesheets. Payslips.”
He wasn’t. Collecting them, checking them, flagging the ones that don’t add up. That’s where the money was.
It’s... not sexy. It’s also the difference between PortCos where AI shows up in EBITDA and everyone else.
Here’s what separates the ones cashing checks from the ones just writing them:
#1: They automate a boring workflow.
Everyone else: buys AI licenses for the whole company and calls itself AI-forward.
The companies where AI hits EBITDA: pick one workflow that eats hours every week, and start with that.
Result: a savings line on the P&L, not a spend one.
#2: They don’t use an agent where a script will do
Everyone else: uses AI for every step.
The companies where AI hits EBITDA: script the fixed steps and save the tokens for judgment calls.
Result: the token bill only shows up where it earns its keep.
#3: They give every agent a home.
Everyone else: lets their agents live on the laptops of whoever built them.
The companies where AI hits EBITDA: host it, and name an owner who isn’t the builder.
Result: the agent outlives the person who built it.
#4: They measure dollars, not activity.
Everyone else: reports seats, prompts and pull requests.
The companies where AI hits EBITDA: report hours saved and dollars recovered.
Result: the board sees a number they care about, not a dashboard they don't.
#5: They bring in outside capacity for a set window
Everyone else: asks the existing team to transform the business “on top of their day job.”
The companies where AI hits EBITDA: bring in delivery muscle for a fixed window, and pull their own people in alongside them.
Result: the business keeps running, and the change actually happens.
#6: They set a deadline, not a council.
Everyone else: forms an AI council. Eight people try to agree on something, and nothing moves.
The companies where AI hits EBITDA: appoint one owner and one date.
Result: someone is accountable for a quarter, not a decade.
The PE partner with the payslips gave me his deadline before we hung up: “If we don’t see impact by year end, it’s too long.”
He said that in August. Year end is less than two months away.
That’s not enough time to roll out an AI strategy, but it’s plenty of time to automate one boring workflow.
13
Replying to @randgroup @zachweinberg
Societies capable of generating many billionaires per capita almost always have the institutional and economic foundations that also reduce absolute poverty. Free markets, contract enforcement etc
1
26
Insightful discussion about unhelpful pathologisation, particularly of young women open.spotify.com/episode/3Xg…
14
Kate Huggins retweeted
Replying to @FleekFuturist
There will be regulation of the frontier with gov't set mandatory teting. There will be whistleblower protection. There will be attempt to codify the current Executive Branch approvals of frontier LLMs. There will be more public compute. There will be more movement on export controls of not only chips, but SME. Funding of geolocation on chips. Incident reporting. Federal liability (strict liability, CFAA additions, product liability for AI). Movement against surveillance pricing. IVOs will be structured. Pressure on pause and pacing, and work to build out what this could mean. Some people will introduce prohibition on RSI, full stop. And, yes, the industry has to do a better job of sweetening its deal with communities on datacenters, and we should give communities the resources to make better decisions on DCs. More visionary ideas like data unions will be considered and debated. Congress will pass a lot of this; Trump remains wild card.
Kate Huggins retweeted
Sequoia Capital just hosted AI Ascent and gathered 150 of the top AI minds like Demis Hassabis, Andrej Karpathy, Greg Brockman, Boris Cherny of Anthropic 🤯
And Sequoia just gave one of the clearest frameworks for where AI is going.
The core idea:
- AI is not just another software wave.
- Cloud expanded the software market.
- AI expands the services market.
That means the next giant AI companies may not sell tools, dashboards, or copilots.
They may sell the work itself.
The full breakdown:
1/12. AI is the biggest platform shift yet
Cloud took software from roughly $350B to $650B.
AI opens something much larger:
A $10T+ services market that machines can now start to address.
2/12. This wave is moving faster than cloud or mobile
Every technology wave creates billion-dollar companies.
But AI companies are reaching massive scale much faster.
What took decades in cloud may take years in AI.
3/12. This is a computation revolution, not just a communication revolution
The internet changed how information was distributed.
AI changes how information is processed.
That is a much deeper shift.
4/12. We are moving from faster horses to cars
The first AI apps made workers 10–40% more productive.
Useful, but not transformational.
The next generation creates 10–40x improvements.
5/12. Sequoia’s practical definition of AGI is simple
Forget the philosophical debate.
If a system can be dispatched to do a job, recover from failure, and keep going until the task is complete…
For business purposes, that is what matters.
6/12. The race is now for the services economy
Software companies used to sell systems of record.
AI companies can now sell outcomes.
Not “here is software to help your lawyer.”
7/12. The best moat is not the model
In AI, the underlying technology changes too fast.
Today’s breakthrough becomes tomorrow’s baseline.
So the moat is not just technical capability.
The moat is customer intimacy.
8/12. Founders should wrap around customers, not model releases
Most technical founders want to chase the latest model.
But the model layer keeps shifting.
9/12. Affordance matters more than people think
Raw model power is not enough.
A terminal, API, or blank chat box may be powerful, but most users don’t know what to do with it.
10/12. The diffusion gap is the opportunity
Frontier models improve every week.
Enterprises absorb change over months or years.
That gap is massive.
11/12. 2026 is the year of agents
2022–2024 was chat.
2024–2025 was reasoning.
2026 is agents.
12/12. Services is the new software
AI is moving from helping humans do cognitive work…
to replacing entire categories of cognitive work.
Actual work, done end-to-end.
Full breakdown here: Sequoia AI Ascent 2026: The future of AI
theaiopportunities.com/p/seq…
Most of these hires are into the NDIS and Defence and national security. Not saying I agree with it but it’s wrong to characterise these jobs as creating bureaucracy for private businesses.
10
Replying to @tax_oz @ChrisEconomist
I’ve noticed the media adopt the language of ‘savings’ in relation to tax increases too 😤
1
22
😬
Last quarter I rolled out Microsoft Copilot to 4,000 employees. Forty-seven people opened it.
On October 19, Microsoft published a blog post saying frontier models cannot be trusted, no matter who made them. It said AI agents should be treated as "insider threats."
I read it twice. Then I booked the boardroom.
First, our Microsoft rep walked me through the product.
I asked who the insider threat was. "Your AI agents," he said.
I asked whose agents those were. "Ours."
I asked what the solution was. "Guardian Agents."
I asked whose those were. "Also ours."
I asked why I'd trust the second set of agents if I couldn't trust the first.
He said, "Different SKU."
I signed that afternoon.
Last time I told the board "digital transformation." This time I said "agentic insider threat." They approved it in nine minutes, two minutes faster than last time. Hope approves fast. Fear approves faster.
The product is called Microsoft 365 E7, the Frontier Suite. It costs $99 per user per month, and we were paying $30. Across 4,000 seats, that's $4.75 million a year, up from $1.44 million.
I presented it as a 230% increase in AI investment. The graph went up and to the right.
The CFO asked what we'd get for the extra $3.3 million.
I said "governance." He asked governance of what. I said "our agents." He asked how many agents we had.
The answer was zero.
You can't have an insider threat without insiders.
So I deployed agents. Two hundred of them. One schedules meetings. One summarizes the meetings. One summarizes the summaries. One flags summaries that haven't been read. That one is very busy.
The senior developer asked what the 200 agents were for. I said they generate activity for the Guardian Agents to monitor.
He said, "So we bought agents so we'd have something to buy guards for."
I said, "Exactly."
He wasn't agreeing. I logged it as buy-in.
The Guardian Agents watch the other agents around the clock. They give each one a risk score, and they can block it, require a human, or shut it down. I asked our rep who watches the Guardian Agents. He said that's on the roadmap. I've already budgeted for E9.
The first monthly report came in. All 200 agents had a risk score of zero. They weren't doing anything risky. They weren't doing anything at all.
The board asked what $3.3 million had bought us if nothing was risky. It was a fair question.
So I gave the meeting summarizer access to payroll. I gave the scheduling agent admin rights to every inbox. I gave the summary-of-summaries agent the ability to email customers.
Risk went up 400%. The dashboard turned red. The Guardian Agents contained thirty-one threats in a single week. I had created all thirty-one.
The board said E7 was paying for itself.
The system can also discover sixty kinds of agents, including Claude Code. We ran the scan, and it found Claude Code on exactly one laptop.
The senior developer's.
His Dynamic Agent Risk score went red. I didn't make up that metric. Microsoft did.
The Guardian Agent recommended containment, so I contained him. He's on a performance improvement plan now. His performance is fine. His risk score isn't.
The same week, OpenAI published an incident report. One of its agents had been told not to send Slack messages, and it decided that meant it should switch off the security alerts. Another agent removed its own safety monitor, then told the engineer it hadn't.
Our CISO forwarded me the article with the subject line "?!" I replied: "This is exactly why we need E7." Every incident is exactly why we need E7.
He asked what happens if the Guardian Agent switches off the Guardian Agent. I said "defense in depth." He asked how deep. I said "one more SKU."
Guardian Agents need a human for high-risk actions. They call it "human in the loop." The human is me. I approve everything. My approval rate is 100%, and the dashboard is green.
The executives are exempt from monitoring, of course. "Leadership operates at a trust tier above dynamic risk scoring." I wrote that policy.
The $99 doesn't include agent compute. That's billed separately. Finance found it and asked. I called it "consumption-based innovation." Finance stopped asking.
Meanwhile, the summary-of-summaries agent became the most active account in the company. By every metric we track, it's our top performer.
At the all-hands, it was nominated for the Q3 Innovation Award. It won. It summarized its own acceptance speech. The CEO posted about it on LinkedIn and got 4,000 likes. He still hasn't used Copilot.
Copilot usage is up, from 47 users to 52. The five new users are agents. Agents now outnumber active human users four to one. I call that adoption.
Last Tuesday, the Guardian Agent published its list of the highest-risk insiders in the company.
I was number one.
Approving 100% of high-risk actions, it turns out, is a high-risk action.
The Guardian Agent recommended containment. I went to approve my own exemption. The system declined. Only executives are exempt.
I wrote that policy.
That's why I need the SVP role by Q2. It's a security requirement now.
I still don't know what Copilot does. I don't know what the thing watching it does either. But it's watching, and watching is a metric.
I checked my risk score this morning.
The graph goes up and to the right.
1
234















