Today's Machine Readable entry is a collaboration between Nat Bullard and Will Hakim, Staff Software Engineer, Halcyon

If you’ve been paying attention to large language model (LLM) benchmarks, the past few months have been as astounding as any since the debut of GPT-3 back in 2022: OpenAI’s o3 producing a step-function improvement on ARC-AGI; DeepSeek’s R1 catching up to – and in some cases, surpassing – OpenAI’s o1 at a fraction of the cost; and Anthropic’s steady drumbeat of impressive model announcements, mostly recently Claude 3.7 Sonnet. And yet, at the application layer, it doesn’t feel like as much has changed.

To use a firsthand example, while LLMs have made massive strides in code generation, they still require a software engineer to conceptualize what they’re trying to build and to articulate that concept in a written text prompt. This disconnect between the advances in foundational technology and the utility of its applications have led us to reflect on past technological transformations and how companies throughout the AI ecosystem should think about positioning themselves to create value in the midst of such a shift.

Our conviction is that AI will be widely available, constantly improving, hard to differentiate at its basic level, and priced competitively and according to specific demand. Equally importantly, it will be part of a process of transforming something of lower value into something else of higher value. And within that transformation, it raises a question: to whom does value accrue in this transformation process?

To explore this idea, we batted around three possibilities about how AI with these attributes could manifest in business. 

The new loom

A concept predating the digital era: is AI the new power loom? The power loom changed a craft - the making of fabric, often at home, using a device but in a highly personal fashion - into an industry. The power loom transformed what had been a home activity and a small financial undertaking into something done at sufficient scale for the creation of mass-manufactured sales and exports.

A change of scale is also a change of scope. An industrial loom didn’t just do the work of many handloom operators - it did it more consistently, and without the same dexterity required. Labor did not go away, but it became different - less about skill in creating a product, and more about skill in maintaining the machine. 

In this vision for AI, the machine / process is extremely powerful, but it delivers very little value to the operator while sending a great deal to the owner. The owner, too, is quite literally an owner as well, of a physical machine and all of the apparatus needed to energize and sustain it. But, value accrues too to those who use not just the process (the loom) but the output (the fabric) and even further, to those who use that output as their own input (apparel makers).  

The new spreadsheet 

Another concept we debated: is business-focused artificial intelligence the next spreadsheet? Spreadsheets were the killer app for an initial subset of customers (finance and accounting professionals) and also a (if not the) killer app for the Apple II when VisiCalc, the first spreadsheet for a personal computer, was released in 1979. 

The spreadsheet was born with one use case and one target market, but it has since become a framework for all sorts of things. Finance, obviously, is still paramount, but spreadsheets are also used for tasks as basic as lists or schedules. It would be hard to say that a spreadsheet is the killer app for a computer today, even if it is essential to workflows across multiple industries. 

But where does the value in being essential lie? Killer app status was not enough for VisiCalc to persist. Microsoft Excel, introduced in 1985, is still with us today as part of Microsoft’s $77 billion annual revenue Productivity and Business Processes business line. But, that business is now smaller than Microsoft’s cloud business. And, it has a direct (and free) competitor in the form of Google Sheets.

We think of its value then, in two ways. The first is in the value it creates for its creator (Microsoft) which is tens of billions of dollars of annual recurring revenue. The second is in the value that it enables for its customers. What has been more valuable: encoding the first =SUMIF() capability, or building something with it? To put it in revenue terms: has Microsoft made more money since 1985 selling Excel, than the world’s financial and professional services firms have earned using it in the same time? 

The new smartphone

Another idea: is AI the new smartphone? Mobile phones existed for decades and sold in the billions before the iPhone arrived and transformed mobile telephony. Prior to 2007, mobile phones (such as Blackberry Messenger) could download and operate software (or, ringtones), take photos, send and receive emails, and operate social networks. But it was the release of the iPhone and App Store, and the subsequent explosion of equally capable devices, that created a universe of new businesses and new values.  

Ubiquitous cameras enabled companies such as Instagram. Device-level, personal geolocation enabled companies as prosaic as Foursquare and as disruptive as Uber. A galaxy of hardware (not just chipsets and cameras but antennas and accelerometers) came together to enable a universe of new businesses built on the combination and concert of other technologies.  

In this imagination, AI is the coordinating layer for all sorts of disparate activities in one frame. Much like an app store, it is a place where people deliver services, ranging from extremely tailored to nearly universal. Much like a device in one’s pocket, it is always on and always available. It is the interface between highly specialized hardware and often-specific software and us, but even still it is only one layer of many in value creation. 

The new force multiplier

These historical examples show how transformative applications amplify existing human activities to unprecedented speed and scope. The power loom didn't invent textile production — it made it possible at industrial scale, transforming what was once artisanal into something that could clothe nations. Spreadsheets didn't create financial analysis — they democratized it, allowing complex modeling that previously required teams of accountants. Smartphones didn't invent photography or transportation — they made them instantly accessible, turning occasional activities into daily habits. In each case, the technology served as a force multiplier for established human needs, creating exponential value not by inventing new activities but by removing friction from existing ones.

That's the promise we’re pursuing at Halcyon: not to replace human expertise, but to dramatically amplify it, allowing energy professionals to achieve outcomes that would have been impractical or impossible before. The most enduring value comes from this amplification — not from incremental improvements to the underlying technology itself.

Comments or questions? We’d love to hear from you - sayhi@halcyon.eco, or find us on LinkedIn and Twitter