British Gas Owner Centrica Cuts 1,300 Jobs as CEO Says Customers Prefer AI Chatbots

Centrica framed 800 of 1,300 cuts around a "targeted deployment of AI tools" while CEO Chris O'Shea insisted customers prefer chatbots and the company denied AI is the cause; a sworn Meta court filing, first reported today, gives the first detailed account of how the company picked its May layoff targets and denies AI chose anyone, while leaving one AI question conspicuously open; and Microsoft's fiscal Q4 earnings Wednesday are the next test for the Xbox reset's 1,600 remaining cuts.

July 2026 so far (through July 27): roughly 14,020 precisely counted layoffs across 28 companies, unchanged from Sunday. Centrica is an energy utility flagged like Allianz and WPP rather than a tech-corporate tally entry, and Crisol Studio's number is undisclosed, so the confirmed count does not move. Several hundred more at Disney/Pixar, about 100 at Native Instruments, and more than 100 at Nothing remain reported as ranges outside the precise tally.

Centrica: 1,300 contact-centre and back-office jobs, and the "customers prefer chatbots" gap

Centrica, the FTSE 100 owner of British Gas, said Thursday July 23 it will eliminate about 1,300 roles over the next two years: 500 call-centre jobs it confirmed in June plus 800 more it now ties to a "targeted deployment of AI tools." The cuts hit customer service and administrative teams in Glasgow, Edinburgh, Cardiff, Leicester, Stockport and Leeds, a roughly 14 percent reduction of the customer operations workforce. Some roles will go through natural attrition, the rest through redundancy. (The Guardian, July 23; Personnel Today, July 23)

The stated reason and the real mechanism are doing two different things, which is exactly the tension this feed tracks. CEO Chris O'Shea told reporters "AI isn't driving these particular job reductions; that's mainly due to changing customer behaviour," pointing to more than 90 percent of customers starting on digital channels and a 20 percent drop in customer calls. (LSE/Reuters, July 23) Yet The Guardian's headline framing, drawn from O'Shea's own claim, is that "most households would rather speak with an AI chatbot than deal with the company's staff." (The Guardian, July 23) The company simultaneously says AI is not the cause and that customers prefer the AI channel. Inbound contact volumes have fallen about 31 percent since 2023, and nine in ten interactions now run through the app, website, chatbot or WhatsApp. (GB News, July 23)

The GMB union reads it as direct substitution. National secretary Charlotte Brumpton-Childs called it "an absolute disgrace British Gas is slashing hundreds of human jobs and giving them to chatbots," saying staff are "being replaced by artificial intelligence." A Centrica spokesman responded that "the suggestion these roles are being replaced by AI or new chatbot technology is simply wrong." (Business Live, July 24; GB News, July 23)

The plausible real mechanism is a contact-centre volume collapse that AI chatbots are absorbing, layered onto a profit decline. Centrica's adjusted operating profit fell 9.5 percent to roughly 497 million pounds in the first half, and the shares dropped as much as 10 percent on the results. The company is putting 92 million pounds into technology as part of its "Customer Service Transformation." (AOL, July 23) O'Shea's own in-house counter-signal: Centrica says it is hiring 500 apprentices this year and recruiting more engineers even as it trims phone staff, the same cut-here-hire-there shape seen at Uber and Verizon this month. Centrica is an energy utility, not pure tech, flagged here the way Allianz and WPP were; it earns its place because the AI-versus-customer-behaviour framing is the clearest version of that dodge since Allianz's CEO named AI outright. Severance specifics were not disclosed.

Meta's sworn filing: managers, not AI, picked the targets, but did AI score the inputs?

A 13-page sworn declaration filed by Meta gives the first detailed, under-oath account of how the company selected workers for its May layoff of roughly 8,000 people, the cut at the center of the lawsuit this feed has tracked since July 15. The declaration, submitted by Linh Doan, Meta's Director of HR Business Partner Enablement, was filed after Judge William Orrick ordered Meta to explain why four plaintiffs on company-sponsored visas were selected, as covered in the Meta AI-layoff ruling. (IBTimes UK, July 27)

Meta's central denial is blunt: AI was not used to determine who would be laid off. Doan describes a human process. Business leaders first identified which parts of the organisation would be affected, then grouped employees into cohorts by job level and role, then established "objective selection criteria" tied to business needs. The criteria had to be finalised before any individual was assessed, and managers were not allowed to depart from them once approved. She also states that Meta's employee activity monitoring tool launched on April 22, after the selection process was already under way, with redundancy notices issued on May 20. (IBTimes UK, July 27)

The filing opens a rare window into Meta's performance system, and it is where the AI question sneaks back in. Meta uses seven performance ratings. One plaintiff received a "Consistently Met Expectations" rating, which the declaration describes as the lowest rating that still meets expectations; despite meeting expectations, the employee was selected because Meta had decided to reduce headcount at that level while retaining people with stronger historical performance. The declaration also calls "Met Most Expectations" a below-expectations rating internally, wording that differs from how many employees would read it. Selection criteria could include job level, tenure, location, specialised skills and organisational structure; managers with fewer than seven direct reports could be selected where the business wanted fewer management layers, with one example reaching six reporting levels below Mark Zuckerberg. (IBTimes UK, July 27)

The leave dispute is where the case stays alive. Doan says managers making redundancy decisions did not know whether an employee had taken leave, and that leave, disability and accommodation status were not in the criteria. But one plaintiff was dismissed less than two weeks after returning from maternity leave; Meta says nearly 40 percent of that employee's team was also selected, arguing she was not singled out. The lawsuit alleges AI usage dashboards did not reflect authorised leave, so recorded activity fell for people who were away. (IBTimes UK, July 27)

The sharpest gap is the one the declaration does not close. It does not say whether employees' use of AI tools influenced the performance ratings that then fed the layoff criteria. The lawsuit alleges AI adoption became a workplace metric through Meta's "Checkpoint programme," and that low AI-tool usage dragged down recorded activity. Meta denies AI made the decisions but has not publicly addressed those specific AI-metric allegations inside the sworn filing. That is the difference between "AI did not pick who was fired" and "AI may have shaped the scores that picked who was fired," and it is the thread the August 24 preliminary injunction hearing will pull with evidence. This is not a new layoff; it is the first sworn look at the mechanism behind the May cut, and it leaves the mechanism's most interesting joint unset.

Briefs

Crisol Studio lets go of its entire team. Crisol: Theater of Idols developer Crisol Studio laid off its whole workforce less than six months after the game launched, warning that full closure is a "feasible possibility." The studio did not specify a headcount but described it as everyone, suggesting a small team. The cause is commercial underperformance, not AI: post-launch momentum stalled and the player numbers could not sustain development, the same commercial-failure bucket as ZA/UM, 1047 Games and OtherSide earlier this month. The game remains on sale and the situation could change with new funding or a publishing deal. (Games.gg, July 27)

Tracking

Microsoft fiscal Q4 earnings, Wednesday July 29 after the close (2:30 PM Pacific), with Satya Nadella and CFO Amy Hood. Consensus is roughly $4.24 EPS on about $87.6 billion in revenue, per TipRanks. It is the first print since the July 6 cut of 4,800 roles, most of them in Xbox, and Xbox content and services are guided down low-teens. About 1,600 of the 3,200 Xbox cuts are still scheduled through fiscal 2027, so the call is the next real catalyst for the feed's most-clicked storyline. (Microsoft)

Xbox WARN separations, September 4. The WARN-effective dates for the covered Xbox reset, including id Software's 136 Texas roles, ZeniMax Online's 213 Maryland roles, Obsidian's 52 California roles and 605 Washington roles company-wide, all land September 4. The ZeniMax Online leadership exodus covered July 16 sits on top of these numbers.

Meta preliminary injunction hearing, August 24. Judge Orrick will revisit the AI question with evidence. The Doan declaration filed this week is the first input, and the unresolved thread on whether AI-tool use shaped the performance ratings that drove selection is the one to watch. Separations from the May cut became final July 22; the four visa plaintiffs face a roughly 60-day clock.

The macro baseline this feed tracks holds: U.S. technology firms have cut close to 140,000 jobs since the start of 2026 per Financial Times and Challenger data, with Amazon, Oracle, Meta and Microsoft accounting for about 50,000 of them, and companies that blamed AI for layoffs have trailed the Nasdaq by roughly 10 percent in the month after their announcements. (TechCrunch, July 25) Anthropic and OpenAI continue to hire engineers and researchers aggressively, the standing counter-signal against a pure-substitution reading.