- Essays··8 min read
Twelve Thousand at Shoprite, No Baseline Underneath
Shoprite runs generative-AI tools across 12,000 employees — 6.9% of its 174,000-person workforce — through a proprietary routing gateway that dispatches each request to the least expensive capable model. The group has not published a pre-deployment baseline, a cycle-time figure, or a per-engineer throughput measure against which the programme can be audited. METR's second-round productivity data shows why that gap matters: the counterfactual is disappearing faster than the measurement apparatus can catch it.
enterprise-airetail-aiworkforce-productivityai-measurementRead - Essays··6 min read
You can buy GPUs but not memory Memory 🙈
The constraint on AI infrastructure in 2026 is not compute but memory bandwidth adjacent to the silicon, and the CoWoS packaging required to deliver it. TSMC's advanced packaging capacity is fully booked through 2027. Adding GPUs does not reduce LLM inference latency when the bottleneck is HBM bandwidth.
ai-infrastructuregpu-memoryhbmhardware-supply-chainRead - Essays··8 min read
The Invoice That Retires the Chatbot
Enterprises are absorbing two simultaneous shocks in 2026: consumption billing that exhausts annual AI budgets months before year-end, and vendor deprecation cycles that retire the very tools those budgets funded. The result is a three-part governance vacuum — an invoice no one can explain, a model no one chose, a shutdown no one knows how to execute — and a single person in the room who approved none of it.
enterprise-aiai-finopsmodel-deprecationagentic-aiRead - Essays··13 min read
The Senior Engineer Whose Week Disappeared Into Somebody Else's Agent
AI coding tools moved the cost unit from headcount to developer behaviour, exposing a gap that finance and engineering share no dashboard to bridge. Individual output climbed whilst throughput stalled: the real bottleneck is senior review time, the constraint the invoice cannot show.
ai-coding-toolsengineering-productivitycode-reviewenterprise-aiRead - Essays··8 min read
The Controls You Cannot Descope
ISO/IEC 42001 certification documents the system an auditor reviewed, not the one running in production today. Working through the standard's 38 controls, each one reveals the same structural gap: a management system written against a fixed architecture, for a product that shipped weekly.
iso-42001eu-ai-actai-governanceenterprise-aiRead - Essays··9 min read
The 73 AI Services Your Network Already Has
Enterprise networks now carry a median of 73 AI services that nobody formally approved — each below procurement thresholds, each handling sensitive data, none linked to a measurable outcome. EU AI Act obligations make deployer liability real for organisations that cannot account for their shadow stack. The argument here is for governance by inventory: a structured three-column sheet that decides which tools stay, which are cut, and what success looks like for the rest.
ai-governanceshadow-itenterprise-aifinopsRead - Essays··8 min read
Emtek's 30 Percent and Indonesia's Ad Curve
Emtek's reported 30 per cent production cost cut is best read against Indonesia's digital-advertising curve — a US$10.7 billion market at 9.4 per cent annual growth — not its streaming subscriber count. The surplus flows into a wider slate and a leaner staffing plan. Against a base rate where 95 per cent of enterprise generative-AI deployments yield no measurable P&L return, Emtek fits the outlier profile — and the risk belongs to any broadcaster copying the deal.
indonesiagenerative-aidigital-advertisingmedia-productionRead - Essays··12 min read
Sunday Essay — What a Southeast Asian Carrier Thinks It Is Buying
When a Southeast Asian carrier commits eight hundred million dollars against a gigawatt of AI factory capacity, it acquires operating control — not architectural independence. The sovereign label sits above a NVIDIA-specified reference design, state-subsidised power, and an enterprise adoption base running at a third of Singapore's rate. The arithmetic depends on year seven, when the subsidy is under fiscal review and the local-language model has commoditised.
ai-infrastructuresovereign-aisoutheast-asiatelecommunicationsRead - Essays··10 min read
The Bill That Ended the Pilot
Uber burned its entire annual AI coding budget in four months — the full allocation gone by spring, no controlled overrun flagged in planning. Three-quarters of enterprises watched AI costs blow past budget last year, and the correction now arriving is triage, not enthusiasm. The governance failure and the cost failure are the same failure: if you cannot name who changed the production prompt, what it scored before it shipped, or what spending ceiling it is running under, the debt is already due.
ai-cost-governanceagentic-aienterprise-aifinopsRead - Essays··8 min read
Where Delhivery's Missing Profit Went
Delhivery's Q1 FY27 results — 55% more parcels, 28% more revenue, 65% less profit — are a precise case study in what Indian logistics companies are not yet accounting for: the accumulating liability of AI pilots whose costs land whilst returns remain perpetually deferred. A small set of applications, route optimisation among them, demonstrably pay back. The rest erode the same margin line the working ones just refilled.
indian-logisticsai-investmentunit-economicsdelhiveryRead - Essays··8 min read
The Agents You Cannot Name
Most enterprises deploying AI agents cannot list what is running, who approved it, or what it costs: a governance gap that is today's operating environment, not a future risk. Agentic workflows trigger ten to twenty LLM calls per task, and token volumes have grown sixty-fold since 2023. The agents you cannot name are the ones that will fail an audit, breach a data policy, or exhaust months of budget because nobody set a spend limit.
ai-governanceagentic-aienterprise-aishadow-aiRead - Essays··11 min read
The Guardrail You Can Audit
The EU AI Act's Article 12 logging requirement is an architectural obligation, not a monitoring one. Guardrails embedded inside the application produce audit evidence the application controls; guardrails deployed as a gateway produce records the application cannot reach. That distinction decides compliance, and organisations retrofitting the wrong architecture will find the fix costs three to five times what building it correctly would have.
eu-ai-actai-guardrailsaudit-loggingcompliance-architectureRead - Essays··10 min read
The Specification You Can Download Is Not the System You Can Buy
A ratified specification is not a commissioned cluster. UALink 2.0 promises 1,024-accelerator pods at half the TCO of NVSwitch, but evaluation hardware is not expected until late 2026, and your purchase order is due in August. The more productive question for enterprise finance is whether InfiniBand or Ethernet closes your deployment faster than a roadmap that has not yet touched silicon.
ai-infrastructuregpu-clustersinterconnectdata-centre-networkingRead - Essays··6 min read
40% Escalated to the Board: The AI Invoice Surprise of 2026
Uber burned its entire 2026 AI coding budget by April, and it is not an outlier: unexpected AI costs changed a business decision at 62% of organisations this year. A charge only reaches directors once it has cleared the spending authority of whoever caused it. Budgets set in 2025, before agentic workloads ran at production scale, are now carrying orphaned agents that burn tokens with no owner and no attributed business outcome.
ai-cost-governancefinopsagentic-aitechnical-debtRead - Essays··9 min read
The $700 Billion Bet Wall Street Just Stopped Believing
Alphabet, Microsoft, Amazon, Meta, and Oracle collectively plan $660–$690 billion in AI infrastructure capex for 2026, yet in July all four companies reporting quarterly results fell on the day despite beating headline metrics. Against that spend, direct AI services generate roughly $25 billion in revenue—a 25:1 ratio that bond and equity markets are now treating as structural risk, not a bridgeable gap. Wall Street has stopped awarding a premium for announced ambition and is demanding contracted margin instead.
ai-infrastructurehyperscalerscapital-expendituremarket-analysisRead - Essays··11 min read
The '19%' problem: what Gartner's productivity surveys actually measured this spring
Two Gartner surveys published in May 2026 reveal a measurement crisis enterprises have not yet named: time saved per AI seat is real, but organisations lack the controls to redeploy that capacity or quality-gate its output before it reaches the P&L. Between 60% and 89% of firms see zero detectable productivity impact despite widespread adoption. That is a governance problem that registers as a measurement problem and gets blamed on the models.
ai-productivityenterprise-aimeasurementgovernanceRead - Essays··8 min read
BMW just put a number on humanoid ROI · the rest of manufacturing is watching
BMW's ten-month, 30,000-unit deployment of Figure AI's humanoid at Plant Spartanburg has shifted the industry's central question from 'can it work?' to 'at what cost, and where?' With unit costs projected below $17,000 by 2030 and current ROI timelines compressing toward 14 months, humanoid robots are becoming a capital-allocation decision. The binding constraint has moved from hardware to integration.
humanoid-robotsmanufacturing-aiindustrial-automationphysical-aiRead - Essays··8 min read
The $500 Million Bill Nobody Anticipated: When AI Slop Debt Becomes a Balance-Sheet Problem
An enterprise client burned $500 million on AI services in a single month, a consequence of deferred governance rather than reckless strategy. AI slop debt, the compounding cost of zombie agents, ungoverned prompts, and abandoned POCs that still bill, has crossed from engineering irritation to balance-sheet liability. When 73% of AI initiatives exceed budget, the board question is no longer which use cases to fund but who can turn off what is already running.
enterprise-aiai-governancefinopsai-agentsRead - Essays··8 min read
Harvey Signs at Raffles Place. The Bill Starts Now.
When WongPartnership signed with Harvey, it acquired a seat licence and a compounding stack with it: wrapper-maintenance costs running 30–60% above the sticker, prompt libraries with no clear owner, and a hallucination-liability perimeter across three jurisdictions that Vietnam's new AI statute just made more expensive. The Forrester 400% ROI ceiling belongs to firms with clean, documented processes. Most Southeast Asian mid-market partnerships are not those firms.
legal-aisingaporeai-governanceprofessional-liabilityRead - Essays··8 min read
Nadella's Half-Drawn Ledger
Satya Nadella's essay on the reverse information paradox names a real problem: enterprise buyers surrender proprietary knowledge to AI vendors with each query: prompts, corrections, evaluation harnesses. The framing is analytically sound. It is also written by the CEO most commercially exposed to a future in which that knowledge lives outside his cloud tenant. The enterprise buyer's job is to notice which parts of the argument survive when the messenger changes.
enterprise-aidata-sovereigntyinformation-paradoxmicrosoftRead - Essays··11 min read
Uber Burned Its Entire 2026 AI Budget in Four Months · The Bill Coming Due
Uber exhausted its entire 2026 AI coding budget by April, four months into the year. Beneath that headline sits a three-layer debt stack: technical debt, ungoverned prompt sprawl, and cost overhang accrued during a subsidised-inference era now ending. Organisations that treat AI spend as an architectural concern from day one will survive what follows; those that audit costs only after the invoice arrives will not.
ai-cost-governanceenterprise-aifinopstechnical-debtRead - Essays··9 min read
Mandates without mastery: the AI upskilling gap that vendor dashboards hide
Corporate AI mandates now outrun demonstrable capability. Google ties usage to performance reviews, Meta tracks AI-assisted output, Microsoft has declared it non-optional. 82% of organisations provide AI training; 59% still report a skills gap. Access is not the deficit. Completion certificates measure the wrong thing, and firms are applying one-time training models to a skill that demands continuous, deliberate practice on realistic tasks.
ai-upskillingenterprise-aiworkforce-developmentskills-gapRead - Essays··7 min read
The Slop Debt Bill Is Due, and Nobody on the Org Chart Owns It
Slop debt is the space between what you pay for AI and what you do not get. The bill is in the inbox; the question is whether the org chart catches up before the auditor does.
slop-debtfinopsagentic-aiai-governanceRead - Essays··8 min read
the kernel of truth and the trillion-dollar bet
A bubble is not the same thing as a lie. There is a real technology, it gets oversold, and the financing built on the overselling becomes the actual fragility. The question that matters: who's holding the debt when demand disappoints?
ai-bubblehyperscaler-capexcircular-financingproductivity-paradoxRead - Essays··7 min read
200 million to Study the Wave You're Making
A frontier lab has priced in the scenario where its product makes office work unnecessary, and is funding the academics who'll help governments respond. Adoption is wide; realisation is narrow. The gap is now numerical.
enterprise-aiproductivity-paradoxworkforceagentic-deploymentRead - Essays··7 min read
the productivity number nobody can actually measure
When the number is real, the baseline comes with it. Everything else is a story about the future wearing a percentage sign.
ai-productivitymeasurementrandomised-trialsenterprise-aiRead - Essays··7 min read
the 0.1% problem · when the productivity number depends on who's holding the ruler
BofA's 10x and the economy's 0.1% measure different things and call both 'productivity'. The number is only as honest as the ruler, and right now almost everyone is holding the ruler that pays them.
ai-productivitymeasurementmetrenterprise-aiRead - Essays··7 min read
The bill nobody booked
The most expensive line item in your AI budget for the next two years is the one your finance team has not yet named. It is sitting in your environment already, in half-built pilots, ghost fine-tunes and redundant copilots, capitalising itself into your monthly cloud invoice.
ai-slop-debtfinopsenterprise-aiai-governanceRead - Essays··7 min read
13%: the upskilling number that quietly breaks the agentic AI business case
Only 13% of employees can actually work with the agents their firms are deploying. The centaur the whole thesis assumes is mostly a novice holding the reins, and AI is demolishing the training ground that produced the experts.
agentic-aiworkforce-upskillingdeliberate-practiceapprenticeshipRead - Essays··11 min read
Sunday Essay · The Three Productivity Curves We Refuse to Reconcile
Three credible measurement traditions, applied to the same workforce in 2026, point in three different directions. The discourse treats them as one number. The gaps between them are the actual managerial signal.
ai-productivityenterprise-aimeasurementcentaur-vs-autopilotRead - Essays··8 min read
The owners are right. That's the part nobody wants to sit with.
The owners are directionally right that they can run with far fewer people. They are very often wrong about which people, how fast, and what it costs to get there. Those two things are the whole story.
ai-and-labourworkforce-displacemententerprise-aiautomation-economicsRead - Essays··7 min read
12% Faster, and Other Numbers I Don't Yet Believe
Four data sets, three methodologies, and no one who matters can yet give you a single defensible figure for AI's effect on knowledge-work productivity. The measurement culture is not up to the tooling, and the vendors will close that gap last.
ai-productivitymeasurement-disciplineenterprise-aicognitive-costRead