Here’s what’s on our plate today:

• 🧪 What Salesforce's hiring freeze reveals about the shifting boundary between human and AI work.

• 📰 The AI layoff powder keg, UK moves to ban under-16s from social media, Zuck sells what Wang builds.

• 💬 Prompt of the Day: Map your team's roles by their exposure to automation.

• 🗳️ Poll: What does Salesforce's hiring line really tell us?

Let’s dive in. No floaties needed…

The Laboratory

TL;DR

• Hiring freeze, one carveout: Salesforce held engineering and G&A flat for roughly 2 years, then kept hiring only in sales: the one function whose job is convincing other companies to replace their own workers with AI.

• Automation's easy targets: The work that goes first is the work documented as a procedure. Salesforce's agents handle 85% of support tickets. Repetition loses before judgment does.

• The line keeps moving: Klarna replaced 700 agents, declared victory, then reversed and rehired them when customers revolted. Every declared safe zone gets revised.

• Experience still has a price: Stanford research shows that AI-exposed roles are shedding 13% of workers aged 22 to 25, while senior staff hold steady. Tacit knowledge still commands a premium.

• The real risk is the pipeline: Entry-level roles are where senior judgment gets built. Automate the bottom rungs, and a decade from now, there's no one left to climb.

What Salesforce's hiring freeze reveals about the shifting boundary between human and AI work

For most of the industrial age, humans told themselves a comforting story about machines. They would take the muscle work, the lifting, the welding, and the sorting, and leave the thinking to them. The line between what a machine did and what a person did seemed natural, running roughly along the border between the body and the mind.

However, with AI, that border is on the move and has now wandered into the office, cutting through the kinds of work humans once filed under judgment. The result is that cognitive labor is entering a phase that, within a few years, will look very different from the one most knowledge workers grew up in, and the evidence that the shift is underway has become difficult to wave away.

As the boundary between human and machine work continues to shift, the more revealing question is no longer whether AI will replace people, but how much of their work it can absorb. To understand that changing balance, it helps to watch a company built on selling automation decide which of its own jobs still require humans.

For most of the industrial age, humans told themselves a comforting story about machines. They would take the muscle work, the lifting, the welding, and the sorting, and leave the thinking to them. The line between what a machine did and what a person did seemed natural, running roughly along the border between the body and the mind.

However, with AI, that border is on the move and has now wandered into the office, cutting through the kinds of work humans once filed under judgment. The result is that cognitive labor is entering a phase that, within a few years, will look very different from the one most knowledge workers grew up in, and the evidence that the shift is underway has become difficult to wave away.

As the boundary between human and machine work continues to shift, the more revealing question is no longer whether AI will replace people, but how much of their work it can absorb. To understand that changing balance, it helps to watch a company built on selling automation decide which of its own jobs still require humans.

The company draws the line

In late May 2026, on a quarterly earnings call, Salesforce chief executive Marc Benioff drew the line out loud. According to Fortune, he told investors the company was "not hiring more engineers" and "not hiring more GA" (general and administrative staff), and was "mostly expanding only in one area," which he placed in "Miguel's area: in sales." By his account, the company's engineering headcount had remained flat for about 2 years, at about 15k people, even as the business grew. The exception, the one function still hiring while the others froze, was the part of the company whose job is to talk to other humans.

Why sales still survive

That exception deserves attention because the obvious explanation is that sales still depend on trust, relationships, and persuasion, qualities that software cannot easily replicate. However, beyond the surface, another reason may explain why the company is drawing the border where it is.

Salesforce sells AI agents that help companies slow hiring in other departments, and those products still require customers to be educated, guided, and convinced. The people Salesforce continues to hire are often the ones who make that happen. In a sense, the company is keeping the workers who persuade other companies to adopt technology that lets them operate with fewer workers. The frontier of human work, at least for now, is drawn partly by who still has to do the convincing.

Trace that logic outward, and the pattern of what gets automated first becomes consistent. The work that goes earliest is the work that can be written down as a procedure, codified, and repeatable. Salesforce shrank its customer support organization as its agents began handling the bulk of incoming questions, and, according to Fortune, those agents now resolve roughly 85% of service inquiries. The same instinct moved through the wider industry over the prior year, from the 14k corporate jobs Amazon cut as it leaned into AI, to the leaner headcount that website builder Wix framed as efficiency. Selling sits on one side of that line, and answering tickets sits on the other, and the gap between them is roughly the gap between improvisation and recitation.

When the boundary moves

However, even now, the border between what can be automated and what requires human judgment is up for debate. Partly, because the trouble with declaring any task safely human is that the declaration keeps getting revised.

Klarna, the Swedish payments company, offers the clearest case of the line moving and then snapping back. It had replaced the work of 700 customer service agents with AI and built much of its public story around the feat, until the experience curdled. According to Bloomberg, chief executive Sebastian Siemiatkowski reversed course and began recruiting people again, conceding that his pursuit of automated cost-cutting had "gone too far" and that customers needed to know a human was available if they wanted one. The company had found the frontier the hard way, by crossing it and feeling the ground give.

From assistance to replacement

If the line is moving, the more uncomfortable finding concerns the direction it moves within a single job. Anthropic, which builds the Claude models, publishes an index of how its systems are actually used, sorting interactions into augmentation, where the software helps a person think, and automation, where it simply does the task. According to Anthropic, augmentation currently leads in its consumer product, at 52% of conversations against 45%, but the longer view shows automation's share slowly climbing, and the company expects tasks to migrate from consumers to businesses as it becomes reliable enough to run without supervision.

Anthropic frames migration as an early indicator of AI's economic impact, since business adoption is where productivity gains first appear. The human-in-the-loop, in this telling, is a phase some tasks pass through on the way to needing no human at all.

This is where the Salesforce exception stops being reassuring and starts behaving like a clock. The boundary between real judgment has already been shifted several times in living memory, each time insisting the new safe zone was different in kind.

Driving was supposed to be too contextual for machines, then chess too creative, then writing too human, and each retreat was narrated as proof of what only people could do, right up until it was not.

Right now, sales sits comfortably on the human side of the boundary for genuine reasons: trust, accountability, and reading the room. Whether those qualities are different in kind or merely the next stretch of road the technology has not yet paved is the question the pattern keeps refusing to settle.

What AI struggles to replace

There is a real counterweight to the alarm, and it deserves to be stated plainly rather than waved away. The work that has held up best is the work that grows out of experience rather than instruction.

According to the Stanford Digital Economy Lab, employment for workers aged 22 to 25 in the most AI-exposed occupations has fallen about 13% since generative AI became widespread, while employment for more experienced workers in those same occupations has held steady or kept growing, with the declines concentrated in roles where AI automates rather than augments. The economists who ran the study describe the technology as good at replacing the codified, book-learned knowledge a new graduate brings, and poor at replacing the tacit knowledge that accrues over a career.

Read one way, that is the case for human judgment as a durable asset, the thing a person spends a working life building, and a model cannot shortcut.

The missing apprenticeship

But when read a step further, it becomes unsettling. The judgment that protects the experienced worker was itself assembled on the lower rungs, in entry-level jobs, fielding basic questions, writing simple code, and sitting in on the call that taught humans how the room works. Those are precisely the rungs that the same research shows thinning first. The human jobs companies still protecting tend to be those at the top of the ladder, even as the rungs below are increasingly automated. That raises a longer-term question: if entry-level and apprenticeship roles disappear, where will the experience and judgment needed for senior roles come from a decade from now? The headlines leave that question unanswered. What they do show is that the boundary between human and machine work is still moving, and it has not yet stopped shifting.

Prompt Of The Day

Tuesday Poll

Bite-Sized Brains

• The AI layoff powder keg: The wave of AI-driven layoffs is building into a volatile political and social flashpoint as workers, unions, and lawmakers push back against automation-led cuts.

• UK moves to ban under-16s from social media: The UK announced plans to bar under-16s from social media platforms, one of the most aggressive moves yet to regulate kids' access to the internet.

• Zuck sells what Wang builds: Meta hired Alexandr Wang to build its AI, and now it's Mark Zuckerberg's job to sell it, a telling split between technical firepower and the pressure to prove returns.

The Toolkit

• Synthesise: No-code platform for building and monetizing AI-powered digital products without writing a line of code.

• Runway: AI video and image generation tool used by filmmakers and marketers to produce cinematic content from prompts.

• Tabnine: AI coding assistant that runs privately on your stack, useful for teams that can't send code to public AI services.

Rate This Edition