Published on LinkedIn and amitabhapte.com on 29th Mar 2026

This week, ambition collided with reality across nearly every front of the AI story. Agentic commerce promised to remove the human from the checkout. A new model leaked before it was ready to launch. The internet crossed a threshold most of us hadn’t noticed. And one logistics giant did something quietly radical: it decided to teach half a million people to work with AI rather than step aside for it. Different signals. Same underlying tension. The gap between what AI can do and what organisations are actually ready for is growing. That gap is where leadership happens.

Agentic Commerce: The First Honest Report Card

Late last year, OpenAI launched Instant Checkout, a feature that lets shoppers complete purchases directly inside ChatGPT without ever visiting a retailer’s website. Walmart signed up as the launch partner. Etsy and Shopify quickly followed. The narrative was compelling: conversational commerce had arrived. The results were not. Walmart has now disclosed that conversion rates inside ChatGPT were three times lower than for click-out experiences that redirected users to Walmart’s own site. That is not a rounding error. It is a structural finding. OpenAI has since moved on, phasing out Instant Checkout in favour of an app-based model that gives retailers more control of the transaction. Walmart is now embedding its own chatbot, Sparky, directly into ChatGPT and Google Gemini, rather than handing the checkout process to a third party.

Meanwhile, Gap has become the first major fashion retailer to launch direct checkout within Google’s Gemini platform, part of an emerging Universal Commerce Protocol that Google has been rolling out since January. The approach is different in intent: Gap is pairing the checkout integration with an AI-powered sizing tool, specifically targeting the return rate problem that plagues online apparel. Net sales were up 2% in Q4 2025, with online sales growing 5%. The CTO was clear that this is about solving real customer problems, not chasing innovation for its own sake.

My PoV: The Walmart data is the most useful signal the agentic commerce story has produced. Consumers will use AI for discovery. They are not yet ready to surrender the checkout to it. The friction of a familiar interface, seeing the full cart, entering payment details on a trusted site, provides reassurance that an embedded AI flow does not yet replicate. For enterprise and retail leaders, the lesson is architectural: own the transaction layer. Let the AI own the discovery. The retailers now embedding their own branded experiences into AI platforms, rather than ceding the whole journey, are making the smarter structural bet.

Anthropic’s Week: Values as a Growth Strategy

It has been a remarkable few months for Anthropic. A public standoff with the US Department of Defense over the use of Claude in lethal autonomous systems, followed by Super Bowl ads that went after OpenAI’s decision to serve ads to its users, has produced something unexpected: a subscriber surge. TechCrunch analysis of 28 million US consumer transactions shows paid subscriptions more than doubling since the start of 2026, with record new sign-ups in January and February. Web traffic was up 43% month-on-month in February and nearly tripled year-on-year. Most new subscribers are at the entry-level Pro tier at $20 per month. Claude Code and Claude Cowork, the developer and productivity tools released in January, have accelerated that growth further.

The same week brought a different kind of Anthropic headline. The company inadvertently exposed an internal draft blog post in a publicly searchable data store, revealing a new model under development called Claude Mythos. The draft described it as the company’s most powerful model to date, part of a new capability tier called Capybara, significantly beyond the current Opus tier. The document also described the model as posing unprecedented cybersecurity risks, specifically for its ability to identify and exploit software vulnerabilities at speed. Cybersecurity stocks fell immediately: CrowdStrike, Palo Alto Networks and Zscaler each dropped around 6%. Anthropic confirmed the model exists and is being tested with early access customers.

And underpinning all of this, a new report from Human Security found that AI and automated traffic have, for the first time, overtaken human traffic on the internet. Automated traffic grew eight times faster than human traffic in 2025. AI-driven traffic alone grew 187% across the calendar year. The internet was built on the assumption that a human being was on the other side of the screen. That assumption is no longer safe.

My PoV: Three separate but connected signals from Anthropic this week. First, values can be a growth driver. Taking a principled public position on how AI should and should not be used attracted consumers in a way that a model benchmark never could. That is worth paying attention to for any enterprise working out how to position itself in the AI market. Second, the Mythos leak reminds us that the cybersecurity stakes are rising with every capability jump. Anthropic’s plan to give cyber defenders early access before general release is the right instinct, but the gap between what AI can do and what defenders are prepared for is narrowing fast. Third, if the majority of internet traffic is now non-human, the infrastructure assumptions of most enterprise digital strategies need revisiting, from fraud detection to API design to web analytics.

AI and the Financial System: A Stress Test No One Planned For

While AI companies attract record investment, a less discussed story is developing in private credit. Shadow banking, the network of private credit funds, business development companies and non-bank lenders that has grown significantly since 2008, has been heavily exposed to software-sector loans. The concern, now surfacing in the mainstream, is that AI may be systematically undermining the value of the software companies these funds have lent against. Apollo Global Management was among the first to flag it publicly last year: “Is software dead?” is now the question private credit managers are trying to price. A closely watched index of 44 business development companies shed around $5 billion in February. The Bank of England has announced it will conduct the world’s first stress test of the shadow banking sector. Lloyd Blankfein has drawn parallels to 2005 and 2006, when hidden leverage was building quietly beneath a rising tide.

My PoV: This story rarely appears in AI newsletters. It should. The thesis is straightforward: if AI can replace significant portions of software development work, the revenue and margins of many mid-market software companies, which form the collateral base for billions in private credit, come under structural pressure. This is not a prediction of an imminent crisis. It is an observation that the financial system has not yet priced AI disruption into the sectors most exposed to it. For enterprise technology leaders, this has a practical implication: the cost, availability and terms of technology financing are likely to become more volatile. AI is not just disrupting products. It is beginning to reprice the capital structures behind them.

The Workforce Question: FedEx Bets on Learning

Against a backdrop of sector layoffs and automation anxiety, FedEx has launched what may be the largest corporate AI upskilling programme in logistics. The initiative covers more than 400,000 employees globally, with personalised, role-based modules that will update monthly. The programme is explicitly tied to internal promotion pathways. The company calls it “promotion-ready” AI training. Frontline workers are already applying for corporate roles at higher rates since the programme launched. Every C-suite executive at FedEx spent two days in Silicon Valley selecting the right technology partners before a single module was deployed.

This matters in context. UPS announced 30,000 layoffs on top of 48,000 the previous year. FedEx has also made cuts. But its strategic posture is distinctly different: use AI to make the workforce more capable, not smaller. The company measures something it calls AIQ, an AI quotient, tracking progress rather than just completion rates. Chief Data and Information Officer Vishal Talwar was direct: “We are measuring progress around AI, not necessarily just success, because it’s going to be very difficult to say this success is only attributed to AI.”

My PoV: Only 28% of organisations have embedded continuous AI learning, according to Accenture’s 2026 Pulse of Change report. FedEx is in that minority, and it is moving at a scale that few others have attempted. The principle behind the programme is the right one: AI literacy cannot be a specialist skill. It needs to be a baseline capability across every level of the organisation, from warehouse floor to boardroom. The harder question is how to measure the business value generated rather than just the learning hours logged. That measurement challenge is the next frontier of enterprise AI investment discipline.

My Takeaway This Weekend

Four stories, one shared theme: AI is not meeting reality gently. Agentic commerce stumbled at the transaction layer, where trust has always mattered most. A powerful new AI model leaked before its makers were ready, and the market reacted to the risk before the product even shipped. The internet crossed a threshold that few enterprise strategies were built for. And in logistics, one company decided to bet on its people rather than against them.

The leaders navigating this well are not those with the most ambitious AI roadmaps. They are the ones who are honest about where friction is real, where trust has not yet been earned, and where their own organisations need to build capability before deploying it. Friction is not a failure of technology. It is the system telling you where the work still needs to be done.