AI + Careers Briefing — November 22, 2026
Today's stories are about the terms of work: who employs you, what you sign away, and what a job asks of you. The Verge reports that New York City's Delivery Protection Act would require last-mile delivery firms operating warehouses to employ their workers directly. TechCrunch reports that microdrama startup Shortical advertises a "license your face" program without saying what it pays. OpenAI's documentation says GPT-6.2 Astra is more concise and direct by default than earlier models, and The Decoder, summarising a New Yorker report, describes psychological pressure inside AI labs.
What would New York's Delivery Protection Act change for delivery workers?
The Delivery Protection Act would require last-mile delivery firms that operate warehouses in New York City to be licensed by the city and to employ their workers directly, according to The Verge.
The Verge reports the bill would effectively ban the third-party arrangements Amazon currently uses for many deliveries, including small companies and independent contractors. Amazon is now fighting it on its shipping labels. The Verge says the company is printing a message on labels that the Council could add up to $5 to any Amazon order. Earlier in the week, a similar notice appeared in Amazon's mobile app and on its website.
What each side says
Amazon's position, in a message to New York customers, is that it relies on a network of small delivery businesses. It says that if the bill passes, it would have to stop working with them and might move operations out of the city. Amazon says the result would likely be slower, more expensive delivery for millions of customers, and that the bill could displace thousands of New Yorkers from their jobs. That page does not explain how Amazon calculated the $5 figure.
amNY reported that the bill, Intro. 0518, is sponsored by Council Member Tiffany Cabán and would restrict large delivery companies from contracting with delivery service providers. It reported that the measure had 35 Council sponsors and was still in the committee on consumer and worker protection.
Cabán told amNY it is "simply dishonest" to claim the bill would require charging consumers $5 per delivery, and she said she does not believe a surcharge would be necessary. Council Speaker Julie Menin called the surcharge a "deceptive campaign," per the same report. The Verge adds that labor unions and Mayor Zohran Mamdani have also pushed back, accusing Amazon of trying to intimidate the Council and acting out of greed.
amNY also quoted people on the other side. Two owners of Amazon delivery service partner businesses, each with nearly 200 employees, said the bill would threaten their companies or push Amazon out of the city.
What this means for job seekers: This is our analysis, not a forecast of how the dispute ends. Both sides describe the same bill and predict opposite outcomes for local jobs, so treat any single headline number as one party's claim.
If you are considering delivery or warehouse work, the practical question is who pays you. Ask the recruiter or the posting to name the legal employer in writing, and whether that is Amazon or a separate delivery company. Ask whether you would be hired as an employee or engaged as a contractor. Under the arrangements the sources describe, the company on the label and the company on your paycheck can differ.
Because the bill's status can change, check current reporting before making decisions that depend on it. Whatever happens, a vague answer about who employs you is a reason to keep asking. For broader tactics on evaluating offers in a hard market, see our guide to the 2026 job search.
Can you really get paid to license your face to AI?
Yes, but the terms are murky: microdrama startup Shortical advertises a "license your face" program for its AI-generated series, yet it has not said what it pays and has not signed any contracts through the program, according to TechCrunch.
Shortical's latest series, "Inevitable," premiered in Los Angeles last week. It stars Aki Avni, an Israeli film and television actor, whose face was used to create a digital avatar for the wholly AI project. A company representative told TechCrunch that the offering is still in its early days and hasn't yet resulted in any signed contracts. The representative said Avni's deal was a direct agreement rather than one made through the platform, and that his compensation was in line with industry standards for a comparable non-AI acting role. The company declined to share a specific figure.
What Shortical has and hasn't said about pay
Shortical's website invites interested people to submit their social media handles, after which the company decides whether to offer a contract setting out the terms, TechCrunch reports. Asked how much it is offering, the company wouldn't say. Asked whether creators could get royalty-style payments like the residual system in television, CEO Guy Shimoni answered that it is handled case by case.
For a sense of the market, TechCrunch cites a Rest of World report finding that microdrama firms in China pay people between US$15 and US$700 to license their facial likenesses. Rest of World separately reports that one platform, ActID, lists per-episode prices of 99 to 500 yuan ($15 to $74) and takes a 10 percent commission. It also quotes lawyer Yile Deng warning that users may effectively lose long-term control over how their biometric identity is used.
TechCrunch also notes that Shortical recently partnered with Payouts.com to build a system for paying creatives, and that some creators have claimed their likenesses were stolen.
What this means for job seekers: This section is our analysis, built on the reporting above. If your work involves your face, voice or on-camera presence, a likeness license is a contract you should read like a job offer, and the reporting suggests a few questions worth asking before you sign.
First, ask for the number in writing. Shortical's refusal to name pay is the clearest warning sign, and our guide to pay transparency and salary negotiation applies here too: a vague "case by case" answer gives you nothing to compare.
Second, ask whether the fee is one-off or tied to ongoing use. Shortical's answer on residuals was noncommittal, so get any royalty promise in the contract itself. Third, ask about scope and duration: which productions, which formats, and for how long, given the lawyer's concern about long-term control.
Finally, treat this as one income stream rather than a career plan. The low end of the range Rest of World reported is $15, and Avni's deal was a direct agreement, not something the open program has yet produced. For broader ways to build durable skills as AI reshapes creative work, see our look at AI-proof career skills.
Sources: TechCrunch · Rest of World
What do OpenAI's GPT-6.2 prompting tips mean for AI-assisted applications?
OpenAI's guidance for GPT-6.2 Astra and Sol says the models write more briefly by default and should act on flagged assumptions instead of stopping to ask, so prompts built for older models need a second look.
OpenAI's developer documentation states it directly: GPT-6.2 Astra is more concise and direct by default than earlier models. The same page advises revisiting instructions that ask a model to be concise when migrating. The Decoder reports that OpenAI released copy-ready prompts for writing style, initiative and coding, and that Astra is built for the most demanding tasks while Sol handles complex work at lower cost.
What changes in how the models behave
According to The Decoder's summary of the guidance, when intent is unclear the models sometimes suggest a next step and wait for confirmation. OpenAI recommends a prompt that has the model carry out authorized steps itself. It should pause only for a major change in task scope, a real blocker, a decision the user must make, or an irreversible or destructive action. When details are unclear, the model should make reversible assumptions, flag them and keep working.
The guidance also addresses final answers. OpenAI's documentation says the final answer should be fully self-contained and written anew for the intended reader, without working shorthand carried over from the model's reasoning.
What a separate evaluation found
A separate evaluation by Artificial Analysis, covered by The Decoder, measured the shift. Astra's average answer length fell from 352 to 210 words, a 40 percent decrease. Sol's fell from 337 to 223 words, down 34 percent.
The same report says the models rank lower on admitting uncertainty. Astra's score on that trait dropped from 96 to 24, and Sol's from 98 to 51. These are relative rankings on a scale from 1 to 100, not percentages. As The Decoder explains, a score of 24 does not mean Astra admits uncertainty in only 24 percent of its answers. The report adds that the personality scores describe communication style, not overall answer quality, and that a lower Transparency Index does not necessarily mean a model gives wrong answers more often.
What this means for job seekers: This section is our analysis, built on the facts above.
First, treat prompts as perishable. If you saved a resume or cover letter prompt that asks for a concise result, the reporting suggests the shorter defaults may now compound that instruction. State the length, structure and level of detail you want, and test the output before you rely on a saved template.
Second, the guidance describes models that make assumptions and flag them rather than ask. For an application, that means reading the flagged assumptions closely. A model filling a gap about your job title, dates or responsibilities is making a guess, and only you can confirm it. Supply those facts in the prompt so there is less to guess.
Third, because the evaluation ranks the models lower on admitting uncertainty, a confident-sounding draft is not evidence that it is accurate. Check every claim, number and date in AI-written application material against your own records before you send it.
For the basics, see our guides on using AI to write an ATS-friendly resume and how to job search in the AI era.
What is pressure inside AI labs doing to the people who work there?
According to the New Yorker's reporting as summarised by The Decoder, writer Sam Kriss argues that enormous psychological pressure inside AI labs helps explain why highly skilled tech workers are drawn to groups like Maple, a Vermont Buddhist monastery.
The Decoder's summary says Maple was founded in 2011 by Soryu Forall. It adds that, per Kriss's report, about two dozen residents live there full time and hundreds more have attended its AI fellowships and pop-up events in the Bay Area. The report says Maple targets burned-out rationalist programmers, AI engineers and data center construction managers.
What are the allegations, and what does Maple say?
These are allegations, not established findings. Former members accuse the group of systematic manipulation and constant surveillance. They say recordings of residents are meant to train an AI modeled on Forall's mind. A mindfulness coach also says participants suffered psychotic breaks and were then removed from the community.
The Decoder reports that Maple's leadership denies almost all of these claims. Forall denies the allegations and says that, looking back, he doesn't believe he made any mistakes. We have not independently verified any of the accounts, and we are not characterizing the group beyond what the source reports.
What does the wider mood in AI look like?
The Decoder's summary of the report describes an intense atmosphere. One well-known AI researcher describes the mood in San Francisco as "religious terror of the future." The piece also points to an AI Impacts survey from 2024 of more than 1,500 AI researchers, in which respondents put the chance of AI causing human extinction or a similarly permanent, severe disempowerment of humanity at 18 percent on average.
Separately, a LessWrong post by Katja Grace about the survey says it received 1580 valid responses and that the average researcher put an 18 percent chance on that outcome.
Not everyone shares the alarm. The Decoder notes that computer scientist Michael I. Jordan pushed back against AI doom rhetoric.
What this means for job seekers: This section is our analysis, built only on the reporting above.
Nothing in the reporting says every AI employer has these dynamics, and the Maple claims remain disputed. But the Kriss argument that pressure inside AI labs is intense is worth treating as a working condition to investigate, just like pay or on-call load. If you're targeting AI roles, assess team culture before you accept.
In interviews, we would ask how the team talks about its mission and whether disagreement with it is welcome. We would ask what a normal week looks like, how workloads are shared, and how the team handles people who need to step back. Ask how recent departures were handled. Listen for specifics rather than slogans.
Treat your own limits as a screening criterion. If the pitch leans on urgency or a sense of destiny more than on concrete work, that is information about fit, not a reason to doubt yourself.
Our guides on job searching in the AI era and interview questions that reveal how hiring managers think cover how to prepare for these conversations.
What to watch
- The status of Intro. 0518 in New York. amNY reported the measure had 35 Council sponsors and was still in the committee on consumer and worker protection.
- Whether Shortical discloses pay or signs contracts under its face-licensing program. TechCrunch reports the company has not said what it pays and has not signed any contracts through it.
- How you check AI-written application material. The Decoder notes the Artificial Analysis personality scores describe communication style, not overall answer quality.
Related Posts

AI + Careers Briefing — November 21, 2026

AI + Careers Briefing — November 20, 2026
