In 1996, Steve Jobs sat down for a PBS documentary called Triumph of the Nerds. Robert X. Cringely was interviewing him. The conversation turned to Apple’s 1979 visit to Xerox PARC, the research lab where the graphical user interface, the mouse, and a long list of other foundational computing concepts had been invented.
Jobs said, in essence: Picasso had a saying — good artists copy, great artists steal — and we have always been shameless about stealing great ideas.
The line has aged better than almost any sentence in the history of business commentary. It has become the unofficial motto of Silicon Valley. It hangs on the walls of startups. It gets quoted in MBA classrooms. It is invoked, usually badly, as justification for almost any unattributed borrowing of someone else’s work.
The line is also, almost everyone who quotes it misses, a deeply incomplete version of a much more interesting framework.
This piece is about what the famous version gets wrong, what the better version actually says, and why the better version is the single most useful framework I can offer for any senior leader trying to navigate AI transformation in 2026.
The Picasso problem
Let’s start with the easy correction.
The attribution to Pablo Picasso is almost certainly apocryphal.
There is no documented Picasso source for “good artists copy, great artists steal.” Multiple researchers have looked. The line does not appear in any published Picasso writing, interview, or documented conversation. It has been attributed to him posthumously, and the attribution has stuck because the line sounds like something a defiantly transgressive twentieth-century master might have said.
The same line has been attributed, at various times, to Igor Stravinsky, William Faulkner, and T.S. Eliot. The genealogy is murky. Different researchers have credited different originators.
The most likely actual source — the earliest documented version of the underlying idea, in publication that can be verified — is T.S. Eliot.
In 1920, Eliot published an essay collection called The Sacred Wood. One of the essays was about the Elizabethan playwright Philip Massinger. In that essay, Eliot wrote something that is recognizably the source of everything that came after:
“Immature poets imitate; mature poets steal; bad poets deface what they take, and good poets make it into something better, or at least something different. The good poet welds his theft into a whole of feeling which is unique, utterly different from that from which it was torn.”
Read it slowly. There is more in those two sentences than in the entire Picasso-attributed simplification.
Eliot is not saying “great artists steal.” Eliot is saying something far more specific. He is saying that the stealing is not the achievement. The achievement is what you do with what you stole.
Bad poets deface what they take.
Good poets make it into something better, or at least something different.
That second clause is the entire heart of the framework. It is the part that the Jobs/Picasso simplification quietly drops. And it is the part that turns the line from a permission slip into a discipline.
The Xerox PARC story, told accurately
To understand what Jobs actually meant when he invoked the (probably misattributed) Picasso quote in 1996, you have to understand the specific story he was telling.
In December 1979, a small Apple team — Jobs and several engineers — were invited to visit Xerox’s Palo Alto Research Center, known as PARC. The visit was part of a deal: Xerox had taken a pre-IPO stake in Apple, and in exchange, Apple got two days of access to the PARC research.
What Apple saw at PARC was, in 1979, almost incomprehensibly advanced.
The Alto workstation. A graphical user interface with overlapping windows. A three-button mouse for pointing and clicking. Bitmap displays where pixels could be turned on and off individually. Networked computing. Object-oriented programming in Smalltalk. Laser printing. Ethernet. WYSIWYG document editing.
Almost every concept that would define personal computing for the next forty years was sitting in that one lab, in working prototype form, with researchers who had no real idea how to turn any of it into a product anyone could buy.
Jobs, by every account of the visit, immediately understood two things.
First, that what he was looking at was the future of computing.
Second, that Xerox did not understand what they had built. The research culture at PARC had produced extraordinary technology and almost zero commercial intuition about what to do with it.
Apple took the concepts.
This is the part of the story everyone tells.
What gets told less often is what Apple did with the concepts they took.
The Xerox mouse had three buttons and cost roughly $300 to manufacture. The hardware was research-grade. It was unsuitable for consumer use at consumer price points. Apple’s industrial design team, led by Jerry Manock and later by Hartmut Esslinger’s Frogdesign, redesigned the mouse from scratch. One button. Simple enough that no user would need a manual. Apple worked with manufacturers to bring the unit cost down to roughly $15. By every meaningful measure — usability, manufacturability, price, durability — Apple’s mouse was a fundamentally different product than what they had seen at PARC.
The Xerox interface concepts were research-quality demonstrations on a $30,000+ Alto workstation. Apple put the same concepts on a Macintosh that retailed for $2,495 in 1984 — making the technology accessible to a market that had been completely unreachable to Xerox.
The Xerox GUI used a complicated, technically elegant system that required Smalltalk and was deeply tied to research infrastructure. Apple stripped it down, simplified it, sacrificed some flexibility for usability, and built QuickDraw — Bill Atkinson’s graphics library. QuickDraw made the Mac UI possible at consumer price points and consumer speeds.
Apple stole the idea.
Then they made it into something better.
That’s the framework Eliot was pointing at in 1920. That’s what Jobs was actually doing when he invoked the Picasso line in 1996. The stealing was the entry condition. The transformation was the achievement.
The Microsoft layer
It is worth telling one more layer of the story, because it is the part that completes the lesson.
In 1985, Microsoft released Windows 1.0. The product was, by almost every honest account, derived directly from what Microsoft had seen of the Macintosh during their collaboration with Apple on Office software. The early versions of Windows looked enough like the Mac interface that Apple sued Microsoft in 1988 for copyright infringement.
The lawsuit dragged on for years. Microsoft ultimately prevailed, partly on the argument that Apple itself had derived the interface from Xerox PARC, and partly on a separate licensing agreement Apple had signed with Microsoft.
The lesson Jobs took from this — and the lesson he was implicitly referencing in the 1996 Triumph of the Nerds interview — was that the entire personal computing interface had been a chain of openly acknowledged thefts. Xerox researchers had taken concepts from earlier human-computer interaction research at SRI and at Doug Engelbart’s lab. Apple had taken from Xerox. Microsoft had taken from Apple. Everyone, at every step, had been “stealing.”
What separated the participants was not whether they stole. Everyone stole. What separated them was what they did with what they stole.
Xerox stole brilliantly from earlier research and produced research-quality demos that never became shippable products. Apple stole from Xerox and produced products that reshaped consumer computing. Microsoft stole from Apple and built the dominant operating system of the personal computing era.
In every case, the stealing was the easy part. The transformation was where the value was created.
Why the simplified version is dangerous
The reason I am writing about this in 2026 is that the simplified Jobs/Picasso version of the framework has, over the past three decades, produced a particular kind of executive behavior that is now actively damaging AI transformations.
The simplified version says: great artists steal.
In the hands of an executive who has not read the Eliot original, this becomes: we can copy what other companies are doing without doing the harder work of transforming it for our context.
I have seen this play out, repeatedly, in AI transformations across multiple industries in 2025 and 2026.
A retail company sees that another retail company has deployed AI-powered customer service successfully. They decide to “steal” the same approach. They copy the architecture, the vendor selection, the use case definition, almost line for line. They deploy it. It fails — because their customer base is structurally different, their existing service infrastructure is structurally different, and the parts of the implementation that they needed to transform for their context were the parts they didn’t transform.
A financial services firm sees that another bank has used AI for credit risk modeling with strong results. They copy the model architecture. They deploy it on their portfolio. It fails — because the underlying data, the regulatory context, and the institutional risk culture are all different, and they didn’t do the transformation work.
A consulting firm sees that another consulting firm has been generating polished AI-augmented deliverables in half the time. They copy the workflow. Their consultants try to use the new approach. The work suffers — because what the leading firm had spent eighteen months building was not the workflow but the judgment about when to use the AI and when not to, and the judgment was the thing that took eighteen months and didn’t transfer in a memo.
Every one of these failures has the same root cause. The executives involved correctly identified what to steal. They incorrectly assumed that the stealing was the work.
Eliot, in 1920, had named the failure mode in advance.
Bad poets deface what they take.
What he meant: bad poets steal effectively but fail to transform. The result is a damaged version of the original — recognizable, derivative, less valuable than the source. Defaced.
The contemporary executive version of this is the AI initiative that copies someone else’s successful pilot, deploys it without context-specific transformation, fails publicly, and produces internal cynicism about AI more broadly. The pilot is defaced. The company is worse off than if they had never attempted it.
The Eliot discipline, translated for 2026
If Eliot is right that the stealing is the entry condition and the transformation is the achievement, the question for senior leaders becomes: what does the transformation actually consist of?
I think it has four components.
One: context-specific adaptation.
What you steal has to be modified for the conditions in which it will operate. Apple modified the Xerox interface for consumer price points. Microsoft modified the Apple interface for IBM PC architecture. Your AI initiative has to modify the source approach for your industry, your data, your culture, your regulatory environment, your customer base, your team’s existing skill set.
The transformation is not optional. It is the entire work. Skipping it produces the defaced version Eliot warned about.
Two: integration into the existing system.
The stolen idea has to be welded into a coherent whole, in Eliot’s language. It cannot sit as a bolted-on appendage. Apple didn’t just add a mouse to the Apple II. They redesigned the entire computing experience around the mouse — the operating system, the application paradigm, the file system, the visual metaphors.
Your AI initiative cannot be a separate program running alongside your existing operations. It has to be welded into how decisions get made, how work flows, how value gets delivered. The integration is the transformation.
Three: aesthetic and human-experience resolution.
This is the part that almost no business framework captures and that Eliot’s poetic version surfaces. The Xerox mouse worked. The Apple mouse felt right. The Xerox GUI was technically sophisticated. The Apple GUI was aesthetically resolved in a way that produced an entirely different human experience.
For AI initiatives, this matters more than executives realize. The technical capability is necessary but not sufficient. The version of the AI tool that your people will actually use is the one where the human experience of using it has been resolved — the prompts are intuitive, the outputs land in the right format, the workflow doesn’t require constant context-switching. The aesthetic and experiential layer is most of the transformation.
Four: the willingness to make it into something different.
Sometimes the version that emerges from honest transformation is better than the source. Sometimes it’s just different — adapted to your context in ways the source never was. Eliot was explicit that either outcome counts. Better, or at least different.
The executive temptation is to claim the better. The actual achievement is often the different — the version that works in your context, even if it would not work as well in the source context. This is fine. This is, in fact, the whole point. Your version is supposed to be different. That’s what transformation is for.
Three practical questions
If you want to actually apply the Eliot framework to your AI transformation in 2026, three questions are worth sitting with.
One: what specifically are you stealing, and from where?
Most enterprises are stealing from everywhere all at once — vendor pitches, McKinsey reports, conference panels, industry case studies, peer benchmarks. The result is a patchwork without coherence. Eliot’s framework demands specificity. Pick the three sources you are stealing from most heavily. Name them. Articulate exactly what you are taking from each.
If you can’t do this, you are not stealing in Eliot’s sense. You are pattern-matching to fragmentary impressions, which produces the defaced version more often than the transformed one.
Two: what is the transformation work that has to happen to make this work for your context?
For each source, articulate the specific gap between the source context and your context. Different industry. Different data. Different regulatory environment. Different customer expectations. Different team capabilities. Different cost structure.
The transformation work is the work of bridging those specific gaps. If you cannot articulate the gaps, you have not yet understood what transformation is required. The pilot will fail in exactly the ways you have not yet examined.
Three: who in your organization is responsible for the transformation work?
This is the question most enterprises skip and the question that determines whether you produce the defaced version or the transformed version.
In most AI initiatives, the people responsible for the project are responsible for the stealing — the vendor selection, the architecture, the deployment plan. The transformation — the context-specific adaptation, the integration, the human-experience resolution, the deliberate differentiation from the source — is implicitly assumed to happen on its own.
It does not happen on its own.
The transformation work has to be staffed, resourced, and led, with the same explicit attention as the technical implementation. The leaders who do this produce successful AI initiatives. The leaders who skip it produce defaced versions.
The closing thought
There is a small irony embedded in this entire piece that I want to surface before I close.
The quote that has propagated through Silicon Valley for thirty years — “good artists copy, great artists steal” — is itself an example of the failure mode it tries to describe.
The quote was stolen from Eliot. It was simplified in the stealing. The most important clause — the one about good poets making it into something better, or at least something different — was dropped.
The resulting version is shorter, punchier, and more memorable. It is also less useful. It is a defaced version of a more powerful framework. It has been quoted, in defaced form, for a century.
That is exactly the failure mode the original framework was warning about.
In 2026, with AI transformation rewriting nearly every assumption about how senior leaders create value, the temptation will be to grab the punchy version of every framework, every approach, every case study, and move quickly. Steal. Deploy. Move on.
The Eliot discipline says this is the path to the defaced version.
The harder path is to do the transformation work. To take what you steal and make it into something better, or at least something different, for your specific context. To weld your theft into a whole that is uniquely yours.
This is slower. It requires more skill. It demands the kind of judgment that doesn’t transfer easily between contexts and that cannot be outsourced to a vendor or a consulting firm.
It is also the only path that produces the version of your AI transformation that actually works.
Steal openly. Steal from everywhere. Don’t pretend to originality you don’t have.
Then do the harder work of transformation.
Make it into something better. Or at least something different.
That is the framework Eliot named in 1920, that Jobs lived in the 1980s, and that almost every senior leader in 2026 needs to internalize before they spend the most important years of their career producing defaced versions of other people’s pilots.
The world has changed. The leaders who notice will be the ones the next decade is built around.
