HubSpot announces layoffs 'not driven by AI-related efficiencies'
So does a projected annual interest bill of more than $6 billion.
Welcome to Skydance.
The Paramount–Warner Bros. Discovery merger closed today. The combined company is estimated to carry more than $80 billion in net debt.
Management is targeting at least $6 billion in annual run-rate synergies within three years.
Roughly the size of one year’s estimated interest bill.
The savings have to ramp up. The interest doesn’t wait.
There are obvious overlaps in technology, real estate, procurement and corporate functions.
The harder question is how much they can remove without weakening the programming and talent that generate the cash.
Cut too little, and reducing leverage gets harder. Cut too much, and you weaken the business that has to carry the debt.
At that point, the balance sheet isn’t supporting the strategy. It’s determining it.
Claude Moves Into Google Workspace: AI That Can Actually Edit Your Work
Anthropic is taking Claude beyond the chatbot window and directly into the productivity apps where people actually get work done.
The company has launched a public beta that brings Claude into Google Docs, Sheets, and Slides, while also allowing users to open and edit those Google files directly from the Claude interface. The integration is available across paid Claude plans, including Enterprise.
From AI suggestions to AI execution
The important change isn't simply that Claude can “read” Google Workspace files. Claude can now make changes inside them.
In Google Docs, Claude can rewrite sections, modify headings, and make larger revisions while preserving the surrounding formatting. Users can review proposed changes before applying them, keeping a human approval step in the workflow.
Google Sheets may be even more interesting. Claude can create formulas, build pivot tables and charts, add new tabs, manipulate data, and perform more complex data-cleaning or analysis tasks before writing the results back into the spreadsheet.
For presentations, Claude can work with existing Google Slides themes and layouts rather than simply generating content that users must manually transfer into a deck. It can also check its output for issues such as overlapping objects or text that is difficult to read.
A two-way workflow
The integration works in both directions.
Users can open a Google Doc, Sheet, or Slides presentation and bring Claude into a sidebar alongside the file. Claude can understand the current document and selected content and then make changes.
Alternatively, users can start in Claude, provide a Google file, or ask Claude to create a new Google document, spreadsheet, or presentation.
That two-way workflow is significant because it changes Claude from an AI assistant that talks about work into an AI agent that can operate on the work itself.
The bigger battle: Claude vs. Gemini
This puts Anthropic directly onto Google's home turf.
Google already has Gemini deeply integrated into Workspace, including Docs and Sheets. Microsoft is taking a similar approach with Copilot across Word, Excel, and PowerPoint.
Anthropic's strategy is different: instead of requiring users to move their work into a separate AI environment, Claude is increasingly becoming an AI layer across the software people already use.
Anthropic has also expanded Claude into Microsoft's productivity applications, meaning the company is positioning Claude across both major enterprise productivity ecosystems.
Why this matters
The biggest implication isn't another AI feature. It's the shift from generative AI to agentic productivity.
The next generation of workplace AI won't simply answer questions, summarize documents, or suggest formulas. It will be expected to take a request such as:
“Clean this dataset, analyze the results, update the spreadsheet, and turn the findings into a presentation.”
—and actually carry out the workflow.
Claude's Google Workspace integration is another step toward that future.
For Google Workspace users, the decision is also becoming less about which productivity suite contains the AI and more about which AI agent can work best with the tools and data they already use.
The competition between Claude, Gemini, and Copilot is therefore moving beyond model benchmarks and chatbot conversations.
The real battleground is the document, the spreadsheet, and the presentation sitting in front of you.
This wasn't the BS, inflate-our-numbers, CEO-puffery line of "We figured out AI, and now we use AI so well that AI was able to replace people- reward us, markets!!"
This seemed like a CEO telling the truth.
Yamini Rangan said plainly that these cuts are not the result of AI-related efficiencies.
HubSpot is changing what it sells, from building software to helping customers grow with AI, and the company is reorganizing around that.
What that seems to mean is fewer management layers, decisions moving closer to the people doing the work, and teams built around customer outcomes instead of product features.
WHY THAT MATTERS
Right now, very few (I'd say zero, but what do I know) companies have actually used AI to replace whole groups of people.
What's really happening is messier and more human.
AI is changing what companies sell and how they need to be set up, and there's a chance that can land hard on middle management. Not exclusively - I think it also impacts new hires. But it just shows how unpredictable this whole thing is.
That's still painful for the people affected.
But it's a more accurate picture than "the robots took the jobs, whoo hoo!!" and accuracy matters when everyone is trying to figure out what this technology is doing to work.
To be fair, this also came about a month after HubSpot's biggest single-day stock drop in 12 years. The pressure was real. But they could have dressed it up as an AI win, and they didn't.
NOW, THE PART THAT MATTERS MORE
Katie Burke, HubSpot's former chief people officer, is running an online gathering for affected HubSpotters this Friday, October 9, at 9am ET. Alumni from companies that are hiring will be there, and everyone who attends gets a list of open roles.
If you're hiring, comment on her post. If you know someone who was affected, send it to them.
Let's take care of these folks - being out of work can be terrifying, and it takes a village, people.
Economists Doubt AI Will Cause a Job Apocalypse
Technology leaders have repeatedly warned that artificial intelligence could eliminate millions of jobs. Many economists, however, remain skeptical. They expect AI to reshape the way people work, alter which skills are valuable, and potentially widen inequality—but they do not expect a future with dramatically fewer jobs.
The skepticism is partly rooted in history. Previous waves of automation generated alarming predictions that failed to materialize. A widely cited 2013 study, for example, estimated that nearly half of U.S. jobs were at risk of automation within a decade. Yet mass unemployment never followed. Nobel laureate Sir Christopher Pissarides, co-founder of the Institute for the Future of Work, notes that repeated forecasts of huge job losses have consistently proved wrong.
Tech Leaders Warn of Major Disruption
Some technology executives remain far more pessimistic. Anthropic CEO Dario Amodei has warned that AI could eliminate half of entry-level office jobs by 2030. Microsoft co-founder Bill Gates has argued that, without appropriate policies, there could be far fewer jobs within a decade.
Research from the Anthropic Institute has also modeled an extreme scenario in which “cognitive” employment falls by one-fifth by 2030. Such a decline could leave roughly one in five white-collar workers—and one in 10 workers overall—unemployed.
Little Evidence of an AI-Driven Employment Collapse
So far, however, the data provide limited evidence of an AI-driven jobs crisis. An OECD survey of 8,000 employers across 13 countries found that only about one in 10 companies using AI had reduced staff, while just 1% said AI was the main reason for doing so.
Employment rates across most OECD economies remain close to record highs, although there are recent signs of weakening. Entry-level hiring in technology-related fields has slowed in several countries, but economists caution that AI is only one possible explanation. Other factors include a long-term rise in unemployment among young graduates, an oversupply of graduates, remote work, and higher interest rates.
In the U.S., where AI adoption is relatively advanced, research from the Yale Budget Lab has found no clear evidence that AI is significantly changing the composition of employment or pushing recent graduates out of work.
What Would It Take for Mass Job Losses?
TD Economics reached a similar conclusion in an October 1 report. For AI to trigger mass unemployment, three conditions would need to occur simultaneously: AI would have to perform tasks reliably without human intervention, automation would need to become cheap enough to justify replacing workers, and companies would need to adopt the technology rapidly.
So far, those conditions have not materialized. One study estimates that fewer than 2% of jobs have more than half of their tasks that could currently be fully automated using AI and existing software.
TD Economics therefore describes an AI “job apocalypse” as a risk scenario rather than the base case. Even in its faster-disruption scenarios, U.S. unemployment would rise by about 0.7 to 1.4 percentage points above baseline forecasts by the early 2030s. The impact could be considerably larger if rapid AI adoption occurred alongside a recession.
The Bigger Risk May Be Inequality
For some economists, the central issue is not whether AI will eliminate most jobs, but who benefits from the technology. Pissarides argues that AI could deepen existing inequalities if a small number of dominant companies capture most of the economic gains.
The emerging evidence therefore points less toward a world without work and more toward a labor market undergoing significant transformation. AI may eliminate some tasks, reduce demand for certain roles, and weaken opportunities for entry-level workers while increasing productivity and creating new forms of employment. The critical economic question may ultimately be not whether AI destroys work, but how its benefits and costs are distributed.
Silicon Valley’s AI Wunderkind Bets on Privacy With Underdog
At 17, self-taught coder Sigil Wen moved to Silicon Valley and landed in an AI hacker house with Andrej Karpathy. He spent his days hacking alongside future AI heavyweights, testing early versions of technologies that would become Claude, Midjourney, GPT-3, and Stable Diffusion.
Now, Wen is betting on a very different vision for AI.
His new startup, Conway Research, has launched an invite-only beta of Underdog, an AI assistant designed around a radical proposition: your assistant shouldn’t need your data to be useful.
Underdog runs its AI models entirely on the user’s own hardware. No cloud inference. No sending your private conversations and files to a remote data center. For now, it runs on Macs and Windows PCs, with Linux, iPhone, and Android versions planned.
Under the hood is Husky, an inference engine Wen built to make local AI faster and more efficient. Underdog currently runs a 27-billion-parameter reasoning model fine-tuned from Qwen3.8-27B — dramatically smaller than the frontier models powering many cloud-based assistants, but capable enough, Wen argues, to handle everyday tasks such as research, shopping, and homework.
That trade-off is the heart of Wen’s pitch: privacy without giving up useful AI.
“Why should using AI require surrendering your private information?”
That question comes from Wen’s AI manifesto, but it also captures the startup’s entire strategy.
An AI assistant that doesn’t need to monetize you
Underdog’s business model may be even more interesting than its technology.
The app will initially be free and won’t rely on advertising. Because inference happens locally, Conway doesn’t have to absorb the enormous cloud-computing costs that come with running a conventional AI assistant.
Instead, Underdog plans to make money when its users spend money.
The assistant can eventually make purchases on a user’s behalf through Stripe, taking a tiny cut of those transactions — closer to an interchange-style model than a traditional SaaS subscription.
That creates a fundamentally different incentive structure.
The more useful Underdog becomes, the more transactions it can facilitate. It doesn’t need to know everything about you so it can sell that information to advertisers. Your data becomes an asset to protect, not a resource to monetize.
And that matters because AI assistants are becoming increasingly intimate.
A genuinely useful assistant could eventually have access to your email, finances, health information, family details, shopping habits, and work. Giving a company that level of access creates a privacy risk that traditional chatbots largely sidestep because they were never designed to act on your behalf across your digital life.
Underdog is betting that people will eventually care about that distinction.
The young founder behind the bet
Wen is hardly a newcomer to Silicon Valley’s AI underground.
Before turning 18, he was already experimenting with cutting-edge models and managed to get GPT-2 running on an Apple Watch. He later worked with Naval Ravikant on Airchat and became a Thiel Fellow, choosing to build instead of attending college.
His network is equally notable. Conway Research counts Andreessen Horowitz, Khosla Ventures, Hummingbird, SV Angel, the Anthology Fund, and a roster of prominent angel investors among its backers.
Patrick Collison, Stripe’s co-founder, is one of the company’s angels — a particularly fitting connection given Underdog’s transaction-based business model.
Wen’s ambition, however, appears more personal than financial.
“I honestly want to build Underdog for myself,” he says. “I’m building a product that I would be proud for my future children to use.”
That may sound idealistic in an industry racing toward ever-larger models and ever-more-invasive assistants.
But Underdog is making a provocative bet: the future of AI assistants may not belong to whoever has the biggest model. It may belong to whoever earns enough trust to be allowed inside your life.