Steal the playbook.
Thought leadership for innovation, digital transformation, future of work and L&D leaders. Best practice, how-tos, listicles and the plays people actually use. Not the news feed.
CelcomDigi is asking Huawei to co-design how its people learn
A three-year memorandum in Petaling Jaya. Pathways in AI, data, 5G and cloud, and certifications at associate, professional and expert level. Albern Murty says advantage follows learning faster than the pace of change. Azmi Ujang, the chief human resources officer, was in the room.
Why Your AI Program Isn't Delivering Business Outcomes
Most work in large organisations is a waste of time, and AI just makes us faster at it. Redesign the work before bolting AI on top of it.
Kier is judging its Oxford programme by what leaders take back
Fifty leaders. A three-day residential at Saïd Business School, then cross-business Growth Groups with Oxford. Stuart Togwell says humans stay at the helm, and the measure is how the learning is applied.
Leading people through AI change: a guide for people leaders
AI rollouts rarely fail on launch day. They stall a fortnight later, when the review still wants the old output. What managers, champions and L&D teams can do about it, from the first task to the first 90 days.
AI training providers in Australia compared: formats, delivery and fit
Universities, institutes, AI boutiques, capability builders and online platforms all sell AI training. They are not doing the same job. A map of who suits what, the formats and cities each publishes, and the questions to put to every provider.
Godrej finished the advanced learning with the champions nearest the work
More than 200 digital champions finished advanced learning in generative AI, enterprise tools, agentic AI and business process transformation. Close to 6,000 people had already built a foundation through AI Ignite. Vijay Balakrishnan says the aim is to democratise AI without decentralising accountability.
Bank of Ireland says almost 7,000 colleagues finished Copilot training
Some colleagues also did specialist learning with UCD and Cambridge Spark. Copilot users save about 30 minutes a day. The same note lists faster chargebacks, a tax tool and a fraud screen, and it does not show that the learning produced them.
Walmart says training and tools are why associates feel more confident with AI
A June survey drew more than 1.7 million responses from more than 2.1 million associates. Office associates and US frontline associates answer different questions, and people who took the OpenAI training reported a 27 percent confidence gain. A store manager in Mexico built an agent and still runs the store.
YuLife wrote the knowledge down before it let people build agents
Jonathan Roomer and Giles Boon-Williams describe a staged rollout at an insurer of about 140 people. Questions came first, then the company's own knowledge, then skills, with agents last. They put weekly use at 94 percent.
WPP asks what the team will do with the space AI creates
Dr Laura Weis treats AI as an amplifier of the system it lands in. She names four uses for the space it creates, and a two-question test for which decisions stay with a person.
The British Red Cross lost the case-note pilot when the story could not be trusted
Simon Hodgkinson describes a case-note pilot that had backing and then began to lose users from about week six. A worker who would not trust the story rewrote what the tool produced, and the dashboard could not say why.
The Distribution Paradox
It's never been easier to get your music heard, but it's never been harder to get anyone to listen.
Scania dropped the mandatory module and ran the literacy as a team conversation
Patrik Hedljung describes seven hours a team takes together, in place of a fifteen-minute e-learning, with more than 2,000 people through a first version and no executive push, inside a company of 65,000.
Uber gave every employee the agent and left the method open
Sumanth Sukumar describes a pilot of 1,000 people, about 80 percent of them outside engineering, and says Uber has not settled how some 30,000 employees will use the agent.
Somerset Bridge protected the hour because the licence was not the problem
Katherine Hawkins-Jones and Ceri Sharples name time as the largest barrier after a survey of more than 220 colleagues, an AI for Leaders programme for 16 executives, and a Bristol estimate of about 5,000 hours that has not been banked.
Cleveland Clinic put every employee through one day and called them caregivers
Toby Cosgrove's chapter, hosted by Cleveland Clinic, follows a student's question in 2006, the move of front-door parking from doctors to patients and a 2011 day when 43,000 employees were called caregivers.
Adobe handed employees a red box before a manager had approved the idea
Mark Randall's December 2014 interview describes Kickbox as a red box, a prepaid card and six checklists an employee finishes before the company funds the idea.
Bosch kept the supplier problem inside the AI Academy
Bosch's 16 April 2026 account, part of its 2025 annual report, follows Juliane Lüder from a controller role through 18 months beside her job, on a supplier-risk forecast she brought with her. Some 100,000 associates have completed Academy training.
Etsy asks the engineer who erred for the account that made sense at the time
John Allspaw's 30 September 2013 account of Etsy's blameless postmortems asks the engineer closest to a failure for the actions, effects, expectations and assumptions, without punishment for the detail.
ING put headquarters staff on mobility and rebuilt around squads
Peter Jacobs and Bart Schlatmann, interviewed in October 2016, describe the June 2015 rebuild at ING Netherlands: headquarters staff reapplied, commercial and IT sat in the same squads, and releases moved from five or six a year to every two or three weeks.
IKEA reskilled call centre co-workers beside the Billie chatbot
Ingka Group's June 2023 account says 8,500 call centre co-workers were reskilled for remote interior design, digital sales and complex enquiries, after Billie resolved about 47 percent of the enquiries it received from 2021 to 2023.
Intuit treats a follow-me-home as watching, not an interview
Intuit's note on Design for Delight puts a follow-me-home ahead of any solution. The visit is for watching real behaviour, and the problem statement comes before a fix.
Knight Capital left retired code on the server that reached the market
The SEC order of 16 October 2013 finds that on 1 August 2012 one of eight servers still held retired Power Peg code, and that server kept sending child orders until Knight lost more than $460 million.
Moderna's AI academy was built for every employee
Dave Johnson's December 2021 account describes an academy with Carnegie Mellon for every employee, on data quality, statistical thinking, machine learning and ethics. The first cohort had not yet started.
iTutorGroup settled charges that its hiring software rejected older applicants
The EEOC's September 2023 settlement follows a lawsuit that said tutor software rejected women aged 55 or older and men aged 60 or older. The relief is a payment, injunctions on that screening, and training if United States hiring resumes.
The Post Office treated Horizon's data as always accurate
Sir Wyn Williams's Volume 1 finds that Post Office staff knew or should have known Legacy Horizon could show losses that were not real, and that the organisation still treated the data as always accurate.
Boeing set aside 737 MAX concerns that later mattered
The House committee's report on the 737 MAX records concerns about MCAS that were discounted inside Boeing, and a chief project engineer who approved a redesign without those facts.
What Microsoft changed when Nadella pushed learn-it-all
Microsoft's account of The Garage describes the learn-it-all shift, a hackathon that replaced the company meeting, and products that started as employee experiments.
What Project Aristotle found besides psychological safety
Google's re:Work guide ranks five dynamics of an effective team, and it warns that a result which failed to appear at Google may still matter somewhere else.
Amazon writes the press release before the build
In November 2006 Werner Vogels described Amazon's Working Backwards process: the press release and the FAQ come before implementation, including when the customer is inside the company.
Air Canada had to pay for its chatbot's bereavement advice
The BC Civil Resolution Tribunal's February 2024 decision in Moffatt v. Air Canada ordered the airline to pay after a customer followed a chatbot on bereavement fares.
What the andon cord asks the person in charge
Virginia Mason watched Toyota workers stop the line in 2002, then built a patient safety alert any employee can file. The case study is specific about the reply that keeps the reports coming.
Klarna put a person back beside the chatbot
In May 2025 Klarna told CX Dive that cost had weighed too heavily on customer service and that a customer should always be able to reach a person. The assistant was still taking two-thirds of the chats.
McDonald's ended the drive-thru test before it became the operation
CNBC's July 2024 account follows a voice-ordering lab from a startup purchase to IBM and a trial in about 100 restaurants. Taco Bell, Wendy's, White Castle and Presto were still short of a full handover.
GPT-4 helped BCG's lower-scoring consultants most
Fabrizio Dell'Acqua and his co-authors found a 43% gain for consultants below the baseline average, and 17% for those above it. On a task GPT-4 handled badly, people using it were 19 percentage points less likely to get the call right.
AI empathy maps skip the interview
Rachel Krause at Nielsen Norman Group shows why the notes have to come from sessions, tickets and visits. A tool can cluster research a team has already collected, and that is the job she gives it.
How Bezos keeps a big company fast
His 2016 shareholder letter names four habits: obsess over the customer, doubt the proxy, take an obvious outside trend, and make reversible calls with about 70% of the information you wish you had.
Bad news gets edited on the way up
Hierarchies reward silence. Amy Edmondson says leaders set the stage, ask for last week's real version and refuse to punish the person who brought it.
You can now prototype the wrong product in an afternoon
A realistic prototype can be built in a day. Nielsen Norman Group calls the risk the fidelity trap: polish makes a flawed idea look ready to ship.
Innovation theatre is the busiest room in the building
Idea challenges, accelerators and incubators fill the calendar. Steve Blank's test is whether anything gets through the sales channel into a user's hands.
Big companies keep buying startups, then smothering them
Acquired startup employees leave at twice the rate of similar hires, and a vesting schedule does not hold the people the buyer wanted.
Stop asking customers what they want
Jakob Nielsen's rule for customer research is to watch the task. Stated preference predicts only about a quarter of whether a design works.
AI agents are about to redraw your org chart
McKinsey's case for an orchestration chart: the companies seeing value from AI agents are the ones redesigning how the work is organised.
AI makes speaking up harder, not easier
Amy Edmondson says the anxiety around where AI is going will wreck psychological safety unless leaders make the uncertainty discussable.
You're freezing innovation spend and still expecting growth
Nearly 60% of executives are freezing or cutting innovation budgets while a third still expect a quarter of revenue from offerings that do not exist yet.
AI transformations run on trust, not reassurance
McKinsey finds trust predicts AI readiness more than another tool rollout. False job promises destroy it the first time roles change.
Jack Dorsey wants to kill middle management. He's half right.
Block's flat-company essay gets the information problem right and skips the people work that managers actually do.
The AI jobs apocalypse hasn't shown up yet
A new NBER paper finds unemployment for recent US graduates didn't spike in summer 2026, so check the evidence before you restructure.
AI is quietly killing the junior job. Where do your future leaders come from?
Stanford's latest payroll data shows entry-level hiring falling in AI-exposed jobs, and the leadership pipeline is what pays for it.
What human oversight requires once a system recommends
Article 14 of the EU AI Act lists what the person assigned to oversight has to be able to do with a high-risk system.
Which failures are worth running on purpose
Amy Edmondson sorts failure into basic, intelligent and complex, and only one of those is the point of a new bet.
What the double diamond asks a team to learn before it builds
The Design Council's four phases start by sitting with the people who have the problem, and they allow a return when the challenge was wrong.
Who keeps an AI incident register, and what goes in it
The OECD's reporting framework separates an incident from a hazard, and it names the fields a log has to hold.
How a team says an AI mistake out loud
Project Aristotle treated admitting a mistake as the test of a team. An AI miss is where that test shows up.
Kill criteria for an innovation portfolio
Rita McGrath's questions for a venture that already has approval, and a review that can stop a bet.
What AI literacy is at work once the login exists
The OECD treats AI literacy as evaluating, working with and using AI. Most workers need that before they need a specialist track.
What an algorithmic impact assessment makes a team write down
Canada's public questionnaire scores an automated decision before it's built, and again before it goes live.
What corporate innovation owes the person doing the work
Implement is the design thinking phase organisations forget, and the queue won't change until someone owns it.
What psychological safety requires when someone challenges a model
Seth and Edmondson on trust fraying when AI enters the work, and the habits that make challenging a model ordinary.
Who is accountable when an agent acts without asking
An agent that can act needs a named owner and a tested shutoff before you widen its scope.