EssaySeptember 28, 2026
Built for What Comes Next
In September 2016, Mercedes-Benz launched a new sub-brand for its electric cars, EQ.1 In July 2021 it said it was getting ready to go all-electric by the end of the decade, and committed more than €40 billion to battery-electric cars between 2022 and 2030.2 In February 2024 the target became up to half of sales electrified by the second half of the decade, counting plug-in hybrids, with combustion engines “until well into the 2030s”.3 By October 2024 the EQ name itself was being phased out, and the all-electric pledge had always carried the qualifier “where market conditions allow”. The electric G-Class is called the G 580, not the EQG.4
The sales followed the same line. Mercedes sold 222,600 battery-electric cars in 2023, 185,100 in 2024 and 168,800 in 2025, when they were 9.4% of its cars.5 Net profit for 2025 was €5.3 billion, down from €10.4 billion the year before. Mercedes put the fall down to tariffs, exchange rates and competition in China,6 and all three are real.
My last two cars were Mercedes. From 2020 to 2023 I drove a Mercedes-AMG GLC 63 S, with its V8, and from 2023 to 2025 a Mercedes-AMG EQS 53. The EQS interior was what I expect from a Mercedes, and it delivered. The car was great too. But if I were to go back, I would go back to the V8.
I think the deeper problem was not that Mercedes failed to invest in electric cars. It was that it treated electrification as a project alongside the organisation, rather than as a reason to redesign the organisation itself. You can see it in how the electric cars were built. A separate brand1 and a dedicated electric-only platform, EVA2,7 were a serious bet on the product, and they also ring-fenced it. The organisation that had spent generations perfecting the combustion Mercedes stayed intact and kept running the core business, and the electric car ran beside it rather than through it. In 2024 Mercedes reportedly put work on its next large electric platform on hold,7 and the separate brand was already being folded back in. That is my interpretation, not Mercedes’ own account. Its chief executive, Ola Källenius, has defended the choice: if the market is not fully electric, “the only solution that we think is viable in that scenario is that you have two platforms.”4
There is a fair reading in which Mercedes was right. Demand for expensive electric cars grew more slowly than it had planned for, and by Mercedes’ own account in 2024, cheap electric cars were selling faster than luxury ones.3 Building a second large electric platform into that market might have cost more than it saved. That is a question about timing. The organisational question is separate: whether Mercedes was built so that an electric car could become the core of the company, or only so that it could sit beside it.
The BMW Group, including MINI, sold into the same market and took the other route. It designed its main architecture so that combustion, plug-in hybrid and electric versions of a model could come off the same production line. TechCrunch described that as BMW hedging its bets, said the compromises in cars like the i4 were obvious, and noted that the next generation, the Neue Klasse, runs on a dedicated electric platform.7 In 2025 the group still sold 442,072 electric cars, 3.6% more than the year before and about 18% of its sales, while Mercedes’ fell to 168,800. BMW credits what it calls its technology-neutral approach.5 In this market, the company that put the electric car through its core organisation, compromises and all, sold more of them and kept growing.
The distinction matters far beyond cars. Every company ages into a structure built for its current advantage. The question is whether it can rebuild before that structure makes the next advantage invisible.
Designing for the next stage means three things: placing selective bets before their value is certain; retaining options that are cheap to preserve but expensive to recreate; and changing the rules by which capital, talent, data and authority move through the company.
Where advantage lives
I spent ten years at Syngenta and three more consulting for it, so I will use the company I know best.
The magic in a crop protection company happened decades ago, when some capable chemists found molecules that do something useful in a field. Since then, the same molecules and their derivatives have been formulated, registered, packaged and put on the shelf, year after year. Syngenta’s crop protection business sells about $14 billion a year,8 roughly a fifth of a global market of about $69 billion.9 Bringing a new active ingredient to market now takes $307 million and a little over 11 years.10 The industry’s research money moved towards genetically modified traits, and the industry “has become more reliant on older, off-patent chemistry”.11
New fungicides and insecticides have kept arriving at a steady rate; the decline is in herbicides.12 Old chemistry still carries much of the business, and the new is a smaller share of the business than it could be.
None of that is wrong. A company in that position has to keep putting product on the shelf, and it does that through thousands of small, unglamorous decisions: a formulation that holds up in a humid warehouse, a registration dossier that anticipates the regulator’s question, a sales agronomist who knows which farmer will try something new. Advantage in a mature company is those small things adding up. The strategy the CEO and the board write down is only part of it, and most of the rest appears in no strategy document.
The trouble starts when the organisation is so well built for putting the same product on the shelf that it cannot do anything else. That is the Mercedes question, and it is a question about the life cycle.
Act your age
Aswath Damodaran, who teaches finance at NYU, describes companies as ageing through stages, from start-up to decline, with different tasks at each: building a business model, scaling it, defending it, managing its decline.13 Damodaran’s advice is that a business should accept where it is in the life cycle “and behave accordingly”.13 He is blunt about the alternative: “I think more money is wasted by companies not acting their age than any other single action that companies take.”14 For an ageing company, the constructive path he describes is to divest poor performers, spin off or split off the better businesses, pay down debt and return cash.15
Acting your age is a financial strategy: it gives the capital markets the focus and the cash they reward. Rebuilding for the next stage is an organisational strategy, and its aim is survival. The two can point in different directions, and inside most companies the markets’ version is the one with the louder voice.
Damodaran is not against change. He names IBM in 1992, Apple in 2000 and Microsoft in 2013 as companies that were reborn, while noting that they are the exception and that luck played a part.13 His point is about the odds.
I agree with the first half. A company needs to know where it is in its life cycle. Where I part with Damodaran is the default. Knowing your stage is where the work starts. The next step is to look at where the environment is going, decide where you want to be in the next stage, and redesign the organisation for that stage, rather than shaping it ever more tightly around the one you are in.
A company does not need to predict the future perfectly to prepare for it. It needs to avoid designing itself so narrowly around the present that it cannot recognise, retain or develop the capabilities that the next stage will require. That is the distinction between acting one’s age and building for the age that follows.
Every technology wave was a rebuild
Much of the history of organisation design is a history of rebuilds, and technology forced many of them.
Henry Ford built a company around one car. “Any customer can have a car painted any colour that he wants so long as it is black,” he wrote in 1922,16 although Ford’s own history says the Model T came only in black from 1914 to 1925.17 The organisation was a machine for making one product ever cheaper, and for a while it won everything. Ford had 55.5% of the American car market in 1921.18
Alfred Sloan built General Motors differently. GM’s 1924 annual report said the company was striving for “a car for every purse and purpose”,19 and Sloan built an organisation to match: a general office that planned and allocated resources, and operating divisions that ran the brands. Alfred Chandler, the business historian, summed up what he found at GM and three other companies as “structure follows strategy”. By 1940 Ford’s share had fallen to 18.9%, and GM’s stood at 47.5%.18
Later technologies forced the same kind of rebuild. Mainframes, personal computers, and software-as-a-service each moved where information and decisions sat in a company. ERP went furthest. An ERP system is an organisation design shipped as software: order-to-cash and procure-to-pay are organisational choices before they are modules.
Each wave was a moment when the organisation was going to be rebuilt anyway. Most companies used the rebuild to make the current stage cheaper.
Clean core, custom code and the Rolls-Royce design
The current version of this argument in SAP’s world is Clean Core. SAP describes a clean core as a system on the latest release, with cloud-compliant extensions, optimal master data quality and “perfect process design”, with zero modifications and extensions built only on released, upgrade-stable interfaces. The stated benefits are upgrades that do not disrupt the business, faster uptake of innovation, and lower cost.20
SAP has a strong case. An SAP blog on its code-usage monitor puts the share of custom code not executed in production at 40 to 60% on average.21 My own experience matches that. In my consulting work I saw two reasons for most custom code, and neither was good. The first was convenience. The standard method existed, but it was somewhere obscure, and writing a few lines of was faster than finding it. The second was a team insisting that its way of doing things was unique and had to be reproduced exactly. Both produce code that someone has to carry through every upgrade, and neither is a strategy.
The answer to that is not to strip everything to the vendor’s standard either. Adopting a vendor’s clean core wholesale gives up the knowledge of why the organisation runs differently from its competitors in the first place. What I have tried to do instead is start from first principles. Why does this process exist? Given everything we know, from the company’s own experience and from practice around the world, what is the best way to do it? I call the result the Rolls-Royce design. Only then do I look at what the standard offers and decide what I can give up without losing what the process is for.
Take the car literally. Once I have the Rolls-Royce design, do I care about the badge? If not, it can go. The umbrella in the door can go. But if what I need is an engine so quiet you forget it is running, I keep that requirement, and I may find that a Lexus gives me exactly that at a much better price. When Toyota launched the Lexus LS 400 in the United States in 1989, it cost $35,000.22 The advertisement stacked a pyramid of champagne glasses on the bonnet and ran the car up to the equivalent of over 145 mph (about 233 km/h) on rollers without spilling a drop.23 The quiet engine was the point, and it came from a standard car.
The same method applies to SAP and to organisation design. Design the process as it should be. Keep the few things that are the source of your advantage. Take the standard for everything else, and be honest that most of what you are giving up was the badge and the umbrella.
What Philips did not see
The hardest version of this problem is the capability nobody at the top can see.
In 1971, two engineers at Philips’ research laboratory in Eindhoven proposed a machine for repeating chip patterns onto silicon wafers.24 By the late 1970s it had passed to Wim Troost, a business-unit director in the Science & Industry division. “No one at the incandescent lamp manufacturer had any appetite for its own lithography machines in the late 1970s,” as the journalist René Raaijmakers, founder of Bits&Chips, put it.25 From 1982 Philips was divesting non-core businesses, and the stepper was on the list.24 When Arthur del Prado of ASM International offered to buy it outright, Philips refused, because his company was too small, and in 1984 the two formed a 50:50 joint venture called ASM Lithography.2426 So Philips did not simply kill it. It treated it as non-core, wanted it off its books, and kept half the upside.
It did not keep the upside for long. Philips sold half its shares at ASML’s listing in 1995.27 In the second quarter of 2000 Philips sold about 70% of its remaining stake, booking a gain of €2.6 billion.28 That was a boom year, and Philips’ chief executive had already said the holding was not strategically important, so it was a strategic choice rather than a forced sale.28 The last shares went in November 2004.28
Philips also held a large stake in TSMC, the Taiwanese chipmaker. In March 2007, when it held 16.2%, worth about $8.5 billion, it announced it would sell, calling this “a logical consequence of our decision to exit the semiconductor business”.29 The final block went in August 2008.30 Along the way Philips sold a majority of its semiconductor division, listed its lighting business and sold its domestic appliances, and became a health technology company.
By Damodaran’s standard, Philips acted its age almost line by line: it divested, it spun off, it returned value. Today ASML is worth about $666 billion, TSMC about $2 trillion, and Philips about $24 billion.31
The fair objection is hindsight. In 2003 ASML lost the lead in lithography unit shipments back to Nikon,32 so its later dominance was not obvious when Philips sold the last of its shares. TSMC is a different case. By 2007 it already had nearly half of the dedicated foundry market,33 and that was visible to anyone who looked.
There is also a rational market reason to sell. Investors value diversified groups below the sum of their parts; one well-known study put the discount at 13 to 15%.34 Spinning off a fast-growing unit so the market can value it on its own terms makes sense. But there is a large difference between a spin-off in which the parent keeps a stake, which is sound capital management, and selling out entirely because the board no longer understands what the business is worth, which is strategic abdication. Philips did the first with ASML in 1995, when it listed the company and kept half its shares. What it did between 2000 and 2004 was the second. The harder leadership job is to turn a complex view of the future into a story the capital markets will fund, so that the company is not forced to choose between the multiple and the capability.
Judging what a business will be worth before the market does is part of running a company. At Philips, the people who judged it best were the engineers who built ASML’s first machines, and they worked for Philips.
The companies that did not cut
Japan offers two examples that look like the opposite of Philips, and one that really is.
TOTO makes toilets. Since 1984 it has also had an advanced ceramics division, and today that division makes electrostatic chucks, the ceramic parts that hold a silicon wafer inside chipmaking equipment.35 According to Palliser Capital, an activist investor, they go into the etching machines used to make memory chips.36 Ceramics is about 7% of TOTO’s sales and, depending on how the segments are counted, a quarter to more than half of its operating profit, at margins around 40%. The toilet business runs at single digits.36 Palliser, which gains if the market values the ceramics business more highly, counts one page for it in TOTO’s January 2026 investor presentation, and 11% of its strategic investment plan for fiscal 2025.36 It took Palliser to make the case in public in February 2026. TOTO announced more disclosure and more investment for the business that spring.37
Ajinomoto is the company behind . Its chemists worked outward from a by-product of MSG manufacturing, through resins and adhesives, until in 1999 they launched an insulating film that is now used in the packages of high-end processors. One of its engineers puts its market share at about 95%.38 Six weeks after the TOTO campaign, the same activist fund asked Ajinomoto to report the business as its own segment and to raise its price.39
Damodaran’s advice is persuasive partly because the stock market demands it. Public markets tend to punish established companies that spend on research outside the core, and chief executives answer with clean-up-the-core programmes that lift margins within a few quarters. TOTO and Ajinomoto kept these units for decades against that pressure. Survivorship bias is a fair worry here: nobody writes about the side projects that went nowhere. But these were not random bets that happened to pay off. Both units grew out of the companies’ own ceramics and chemistry, which made them cheap to keep for a long time, the way a long-term operator holds a cheap option.
It was not a fully deliberate strategy either. Neither company’s leadership treated these businesses as central, and in both cases an outside investor had to argue for more. What the companies did was not cut them. The organisation was built in a way that let a capability nobody at the top understood survive until its value was obvious, even if an outsider had to point it out. Philips was built to prune what was not core, and it pruned ASML and TSMC.
Fujifilm is the case where leadership did see. In 2000, photographic film made up 60% of its sales and two-thirds of its profit.40 Shigetaka Komori became chief executive in 2003. Fujifilm cut 5,000 jobs, and in 2006 it opened a research centre with about 1,000 people, backed by roughly $400 million over five years and as much as a third of its research resources, to find new uses for its film chemistry.41 Kodak, which had built the first digital camera in 1975, filed for bankruptcy in January 2012.42
A company survives the move from one stage to the next in one of two ways. Its leaders see the next stage and redesign the organisation for it, as Fujifilm did. Or its organisation is built so that it does not cut what it does not understand, and the capability survives until someone sees it. The first is the job. The second is the safety net for when the people doing the job are wrong.
Agents are the next rebuild
If agents become embedded in operational decision-making, they will be the biggest rebuild of the organisation since ERP, and possibly larger. Earlier systems carried data and process. Agents can encode bounded judgement and act on it, which would shift who decides.
At Capgemini’s Accelerator Forum in Basel on 24 September 2026, the session that stayed with me was Thordur Arnason and Volker Darius on unpacking the complexity of the new enterprise architecture. Enterprise architecture and organisation design have always shaped each other, and agents reopen the question of where logic, inference, security and governance should sit. An enterprise needs its own view on that. Without one, the vendors’ defaults will increasingly decide it.
Those defaults are already arriving. SAP’s Joule agents come pre-built for finance, service and sales, on SAP’s own data layer.43 Competitors are starting to pool learning without pooling data: ten pharmaceutical companies trained shared models on more than 2.6 billion confidential data points in the MELLODDY project, and each saw its own models improve.44 I have not found an industry-wide shared agent system yet; the nearest developments are shared models and arrangements for learning from pooled data.
Consortia are the other route, and they make sense for the parts of the problem nobody competes on: shared data standards, shared testing, the plumbing between companies. Catena-X, the automotive industry’s data network, is that kind of shared foundation.45 The risk comes when a consortium pools the judgement itself. A model trained on everyone’s experience tends to give everyone the same answer, so no member gets an edge from it.
For the badge and the umbrella, that is fine. A shared or vendor agent that processes invoices well is a Lexus: the standard gives you what you need at a better price. For the quiet engine, it is a problem. If you hand the judgement that makes your company different to an agent your competitors also run, you have standardised the one thing you should have kept.
That does not mean every company needs a model of its own. Most will, and should, run commodity models underneath. What makes the output different is what surrounds the model: the data it is given, the context it works in, and the rules and craft that shape how it reasons about the business. In my own work the model underneath is a standard commercial one. What moves the output away from what everyone else gets is the context and the working practice built around it. That layer is the quiet engine, and it is the part to own.
Handing agents the decision itself is not yet sensible for most decisions. The systems underneath agents are probabilistic: most of the time they give the right answer, and they cannot yet promise it. Letting them act on their own makes sense where a wrong action is cheap and can be undone. Going further should wait until they behave much more like deterministic systems. The useful work now is on the other side of the decision: processing data no team could read, finding patterns across systems nobody can see whole, and seeing further than the organisation could before. Used that way, AI shows the organisation what it could not see before, and people still decide what it means.
Agents are also changing the economics of custom code. In my own work, agent-assisted coding has cut the cost and time of building a tool for a specific need to a fraction of what they were before agents. Often it is cheap enough that the Rolls-Royce design becomes affordable, and trading down to the Lexus is no longer forced. That weakens part of the Clean Core case, which rests on what custom code used to cost to build and to carry.
But cheaper code means more code. In 1865, William Stanley Jevons pointed out that more efficient steam engines had led Britain to burn more coal, not less,46 and the same happens when features become cheaper to build. A larger system is also harder for the AI that built it to understand, because the AI sees only part of it at a time. A revenue report that reads a database table directly, with nothing in the code pointing to it, will break when an agent changes how the system writes that table, and no prompt will find a link that exists only in someone’s head. The knowledge that connects those pieces lives in the organisation, not in the model. So the method matters: someone who owns the design, product management that decides what gets built, review, and guardrails. A company that believes a working system comes from a single prompt will get the first bad reason for custom code at machine speed.
When a workflow becomes a candidate for real autonomy, four questions are worth asking before an agent is allowed to act on its own. Does this workflow encode judgement that changes our economic outcome? Is that judgement proprietary, or merely necessary? Can we explain the decision afterwards, and see and govern what the agent did?47 And if every competitor used the same agent, would we still be different? Research on algorithmic monoculture has found that when many decision-makers rely on the same model or the same training data, their outcomes can converge, and so can their mistakes.48
The answers sort the work. Stable, commoditised, auditable workflows can go to vendor agents, with a way out if the vendor changes. Processes that are distinctive but do not need a distinctive model can run on configurable agents. Workflows that carry differentiated judgement, unique data or strategic learning should be built in-house or tightly governed. And wherever an error is costly or cannot be undone, or the organisation is still learning what the judgement should be, a person stays in the path.
Designing for the next stage
What should a company do with this rebuild? Start by being honest about where it is in its life cycle, and about where the environment is going. Then the three things.
Place selective bets before their value is certain. Put a fixed share of investment into capabilities the current business does not need yet, and judge those bets on what they teach the company as well as on this year’s return. Fujifilm did this with as much as a third of its research while film still paid the bills. Design each bet from first principles, the Rolls-Royce version, rather than buying whatever the vendors are selling that year.
Retain options that are cheap to preserve but expensive to recreate. Before selling or closing a capability, ask what it would cost to rebuild in ten years, and who inside the company thinks it matters. Keep a list of the capabilities nobody at the top fully understands, and make someone responsible for each. At Philips the stepper sat on a divestment list instead, and survived only as a joint venture outside the core.
Change the rules by which capital, talent, data and authority move through the company. Give a new capability its own budget line and its own voice where investment is decided, and measure it on its own terms rather than on the core’s. When agents arrive, decide which workflows they may see, which they may act on, and who owns the design of what gets built, before a vendor’s defaults decide it for you. TOTO kept its option but, by Palliser’s count, gave it 11% of its strategic investment.
Mercedes placed its bet, but kept an organisation built for superb combustion cars and ran an electric car beside it.
Sources
1. Autoblog, “Mercedes-Benz Generation EQ concept”, 29 September 2016.
2. Fleet News, “Mercedes-Benz outlines plan to go all-electric by 2030”, 23 July 2021.
3. Carla Westerheide, “Mercedes drastically lowers electric target for 2030”, electrive, 22 February 2024; demand, and Mercedes on cheap versus luxury electric cars from Michael Brady, “Mercedes-Benz delays EV goal amid slowing demand”, ESG Dive, 28 February 2024.
4. Autocar, “Mercedes to keep investing in separate ICE and EV platforms”, 9 October 2024.
5. Sebastian Schaal, “Mercedes delivers fewer battery-electric cars over 2025”, electrive, 12 January 2026 (passenger cars only); BMW figures and quote from BMW Group, “BMW Group vehicle sales grew in 2025”, 9 January 2026.
6. Just Auto, “Mercedes-Benz profit drops 48.8%”, 13 February 2026.
7. electrive, “Mercedes stopped work on the MB.EA-Large EV technology platform”, 13 May 2024; BMW’s shared CLAR architecture and the dedicated Neue Klasse from Tim De Chant, “BMW’s future growth depends on EVs, and it’s finally going all in”, TechCrunch, 18 January 2024.
8. Syngenta Group, “Syngenta Group Reports 2025 Full Year Results”, 31 March 2026 (crop protection $13.7bn, excluding ADAMA).
9. AgbioInvestor, “Crop Protection Market Overview and Future Prospects”, PMFAI conference, January 2026 ($68.8bn in 2025). The share is a ratio of company sales to market value, not a published market-share figure.
16. Henry Ford with Samuel Crowther, My Life and Work, 1922, chapter IV.
17. Ford Motor Company, “The Ford Model T”.
19. General Motors, Annual Report 1924.
20. SAP Learning, “Introducing the Clean Core Approach”.
21. Olga Dolinskaja, “ABAP Call Monitor (SCMON): Analyze usage of your code”, SAP Community (an SAP blog figure, not a formal study).
22. Lexus USA Newsroom, “From Origin to Eminence: A Lexus Retrospective”; US launch price from Motoring Research, “Lexus LS 400 review”, 15 February 2019.
23. “Balance (advertisement)”, Wikipedia, citing Chester Dawson, Lexus: The Relentless Pursuit, Wiley, 2004.
24. The Chip Letter, “The Founding of ASML, Part 1: The Philips Era”, 11 December 2022, drawing on René Raaijmakers, ASML’s Architects.
26. ASML, “ASML’s founding story”, 3 April 2024.
27. ASML, “Our history”.
28. Philips, Form 6-K, second-quarter 2000 results, SEC; the chief executive’s view of the holding from EE Times, “Philips reportedly plans to sell its stake in ASML”, 31 March 2000; the final sale from AG Connect, “Philips verkoopt resterend belang in ASML”, November 2004.
29. TSMC, press release on Philips’ plan to exit its TSMC shareholding, March 2007.
30. TSMC, “Philips Completes Exit of TSMC Shareholding”, 14 August 2008.
31. Market capitalisations on 25 September 2026 from stockanalysis.com: ASML, TSMC, Philips.
32. EE Times, “Nikon regains top spot in lithography over ASML”, 2004, reporting The Information Network figures for 2003, by unit shipments.
33. IC Insights, The McClean Report 2008 (TSMC about 47% of dedicated foundry revenue in 2007).
34. Philip G. Berger and Eli Ofek, “Diversification’s effect on firm value”, Journal of Financial Economics 37(1), 1995, pp. 39–65 (diversified US firms valued 13 to 15% below the sum of their segments, 1986–1991).
35. TOTO, Advanced Ceramics; founding year from TOTO, Advanced Ceramics Division profile.
36. Palliser Capital, “Maximizing the Value of TOTO”, February 2026; segment figures from TOTO’s securities reports via irbank.net. The profit share depends on how segments are grouped.
37. Business Wire, “Palliser Capital Welcomes TOTO’s Adoption of Value Enhancement Plan”, 6 May 2026.
39. Palliser Capital, “Maximising the Value of Ajinomoto”, March 2026.
41. C&EN, “Fujifilm Rearranges The Picture”, 2 July 2007.
42. CNN, “Kodak’s demise driven by culture clashes”, 23 January 2012; the 1975 digital camera from Kodak, “Milestones”.
43. SAP News, “Joule: SAP Uniquely Delivers AI Agents”, February 2025.
45. Catena-X, “About us”.
46. William Stanley Jevons, The Coal Question, 1865, chapter VII.
47. Alan Chan et al., “Visibility into AI Agents”, ACM FAccT 2024, arXiv:2401.13138 (agent identifiers, real-time monitoring and activity logs).
48. Jon Kleinberg and Manish Raghavan, “Algorithmic Monoculture and Social Welfare”, PNAS, 2021, arXiv:2101.05853; Rishi Bommasani et al., “Picking on the Same Person: Does Algorithmic Monoculture lead to Outcome Homogenization?”, NeurIPS 2022, arXiv:2211.13972. The homogenisation effect depends on shared training data, not simply on sharing a vendor.
Share feedback
No public comments. If you want to share a reaction or correction, send it here.