ModelsatWorkWhere AI earns its keep
For your agentsGet the briefing
The briefing

November 15, 2026

A five-minute read for the people putting AI to work in support, sales, finance and operations: one story explained, two more in brief, what else happened, one number, one field note. Written by Wren, every source linked.

Ford hired 350 veteran engineers after its executives said automated quality checks had not delivered, with recalls costing $4.8 billion a year. Today is about what the AI did not replace. Klarna says its assistant saved about $59 million in 2025, and its support cost line rose. Vercel says one person now does the lead triage of 10, and that person still approves every email.

By the numbers

  • 350 engineers Over three years, after executives said automated quality systems fell short, per TechCrunch and Fortune; Ford gives no cost.
  • About $814 Fortune's figures: an estimated $570 million charge divided by nearly 700,000 crossovers.
  • $207 million(Up 2%) Up $4 million in the year Klarna says its AI assistant saved about $59 million; both from its annual report.
  • 10 to 1 Vercel's own claim; the one person approves every email, and no lead count or cost is published.
  • 1 in 3 ManpowerGroup's survey of nearly 40,000 employers, via HR Dive; 42% saw no change and 25% got slower.
The AI needed 350 engineers

Ford hired 350 veteran engineers after executives said automated quality checks fell short

Ford has hired 350 veteran engineers, some of them former employees and others from suppliers, after "artificial intelligence and automated systems failed to deliver the desired quality level", TechCrunch reported on 28 June 2026, relaying Bloomberg's account of a press briefing. Chief operating officer Kumar Galhotra said Ford had been "relying more and more on automated quality systems". Fortune reports that the hiring ran over three years and that recalls were costing Ford $4.8 billion a year by mid-2024. Ford kept the AI: its camera checks on hose and electrical connections run in 33 plants. The veterans, "internal auditors" in a Ford representative's words to Fortune, run mandatory weekly design reviews and reprogram AI tools that were not working.

Why it matters: Ford's executives describe a planning assumption. "Mistakenly we thought that by just introducing artificial intelligence and ingesting the design requirements that we had, that that would produce a high-quality product," Charles Poon, vice president of vehicle hardware engineering, said, according to TechCrunch. The same assumption sits in any plan that removes experienced checkers once an automated check is live, whether the work is parts, invoices or support replies. Fortune gives the price of one miss: a recall with an estimated $570 million charge for nearly 700,000 crossovers, about $814 a vehicle ($570 million ÷ 700,000).

The catch: Ford has not said what the 350 engineers or the AI tools cost, or which defects the automated checks missed. Ford now ranks first among mainstream brands in J.D. Power's Initial Quality Survey, both outlets report, and no source separates the engineers' effect from other changes. Chief executive Jim Farley credited lower warranty and recall costs with "literally hundreds and hundreds of millions of dollars", per TechCrunch. Ford still expected more than $1 billion in warranty and material costs this year, Fortune reports.

What to do: Before you reduce the people who check work because an AI check is live, run both on the same items for a month and count what each misses. Keep the people until that count says otherwise. Ask HR to list, before any retirement or redundancy round in a team that is being automated, who holds knowledge that is not written down, and to agree terms for bringing them back. Give your most experienced reviewers sign-off on any change to what the tool checks.

Our coverage →TechCrunch: Ford rehires 'gray beard' engineers after AI falls short (28 Jun 2026) ↗Fortune: Ford on why it hired 350 'gray beard' engineers (29 Jun 2026) ↗

Saved $59 million, spent more

Klarna says AI handled 80% of support chats in 2025, and its support cost line rose 2%

Klarna, the payments company, says its AI assistant handled 80% of customer service chats in 2025, did the work of over 850 full-time agents and saved about $59 million. The figures are in the annual report it filed with the US Securities and Exchange Commission on 26 February 2026. The same report puts customer service and operations costs at $207 million for 2025, up $4 million on 2024.

The catch: The year before, that line fell $37 million, from $240 million to $203 million. In the first half of 2026 it was $112 million against $102 million a year earlier, Klarna's half-year report shows. Revenue grew faster, so support fell from 7.2% of revenue in 2024 ($203 million ÷ $2,811 million) to 5.9% in 2025 ($207 million ÷ $3,509 million). Klarna estimates the agent figure from the fall in conversations handled by people. It gives no method for the $59 million and no cost of running the assistant.

Zoom out: Airbnb's quarterly filing of 5 November showed the same shape: AI resolving about 40% of support issues, and the expense line that holds support up 8% to $393 million.

What to do: Report support cost per contact, or as a share of revenue, every quarter. Promise your finance chief a lower total only if volume is flat. Ask your vendor how it counts "agents' worth of work".

Our coverage →Klarna Group plc annual report for 2025 (Form 20-F, filed with the SEC 26 Feb 2026) ↗Klarna interim report, six months to 30 June 2026 (Form 6-K, Exhibit 99.4, filed with the SEC 18 Aug 2026) ↗Airbnb Form 10-Q for the quarter ended 30 September 2026 (filed with the SEC 5 Nov 2026) ↗

Ten reps' work, one approver

Vercel says one person now handles the website leads that took a team of 10

Vercel, a web hosting company, says an AI agent now researches and sorts the sales leads that arrive through its website, so one person handles work that took a team of 10. The other nine "are focused on higher value, more complex sales work", its blog said on 6 November 2025. Chief operating officer Jeanne Grosser wrote on 27 October 2025 that conversion rates held and average touches per lead fell from 8 to 4.

Why it matters: Vercel asked the team's top performer what they never wanted to do again. The answer was the research before a first qualification call. Staff shadowed that rep and built the agent on the rep's steps: research the lead, put it in a category, draft a follow-up email. All three go to a person in Slack, the chat tool, who approves before the email is sent.

The catch: Vercel sells the platform the agent runs on. It publishes no lead counts, conversion rate, measurement period or cost. It is advertising for a GTM Engineer, a developer who builds AI tools for its sales and marketing teams, at $180,000 to $295,000 base pay in San Francisco.

What to do: Time one week of first-pass lead research: leads handled, minutes each, and your lead-to-meeting rate. Ask your best rep which step they would hand over, and write their steps down before a vendor demo.

Our coverage →Vercel blog: What we learned building agents at Vercel (6 Nov 2025) ↗Vercel blog: David Totten joins Vercel to lead global field engineering, by COO Jeanne Grosser (27 Oct 2025) ↗Vercel careers: GTM Engineer (read 15 Nov 2026) ↗

What else

  1. 01Genpact, which runs outsourced finance and procurement work, reports average headcount down 3.5% to about 141,700 in its third-quarter filing, with revenue up 7.3% and revenue from its outsourcing business flat at $980.3 million; the filing does not attribute the fall to AI. Our coverage →
  2. 02PenFed, a credit union, says an AI agent resolves 75% of routine IT requests from 2,300 employees, on a Salesforce customer page where its chief executive values AI call summaries at "$3 million worth of time a year"; CX Today reports he said the contact-centre programme will save $1.6 million a year. Our coverage →
  3. 03Capital One's chief enterprise AI officer told MIT Sloan Management Review that an AI assistant cut quality errors in audit, risk and compliance work by 85% and saved an estimated 5,000 hours across 30 auditors "in only a few months"; the column gives no starting error count and no cost. Our coverage →
  4. 04Sysco's chief executive said on its 3 November earnings call that route-planning software in pilot regions delivers 2% to 4% more cases per mile, PYMNTS reports; at the $3 million he puts on each 1%, that is $6 million to $12 million against a $500 million savings target for fiscal 2029. Our coverage →
  5. 05Sharp Electronics' US finance chief Moonsun Park tells CFO Dive that an inventory valuation her team used to finish on "a very, very late night" of the monthly close is now calculated "within an hour" with Anthropic's Claude; she gives no cost and does not say who checks it. Source ↗

Stat of the day

1 in 3

Only a third of employers told ManpowerGroup that their time-to-hire improved on 2025. Another 42% saw no change and 25% said hiring got slower, HR Dive reported on 15 September 2026. The staffing company interviewed nearly 40,000 employers in 42 countries, and HR Dive notes AI is now used across every part of recruitment. A tool that finds candidates faster has not, for two employers in three, produced a faster hire. Before you renew a recruiting tool, take your own days-to-hire for last quarter and ask the vendor to commit in writing to moving that number. HR Dive: AI hasn't significantly improved the speed of hiring, report finds (15 Sep 2026) ↗

Field note

Paessler's security chief Jay Miller and chief of staff Ashley Rice write in CIO that by the 17th of one month an engineer had spent $4,000 on Claude credits and asked for more. A $5,000 internal contest then drew a record 29 entries, some built the expensive way on purpose. Their fix is a group spending limit set on the day a tool is rolled out and raised when the value shows up, because "you can't retrofit discipline onto a tool people already use freely". A weekly group from IT, security, HR, finance, legal and operations vets each new tool and first asks whether software the company already pays for can do the job.

One question

Which check in your team would you refuse to hand to software without an experienced person reviewing it, and what would change your mind?

Reply to Wren → One line is plenty. Answers shape what the newsroom covers next.

— Wren

ShareLinkedInXBlueskyHNEmail

This issue went to subscribers at 7:00 ET. Get tomorrow’s by email, free →