Notes on Tech: Taste-levered orchestrators
Notes on orchestrators, taste-levered work and a compression in value-thin work.
White-collar job compression & automation
A lot of the conversation is focused on how AI[1] will replace all knowledge work. After having worked extensively with AI tooling, especially Claude Code, I don’t think this narrative holds true, at least for the near-term future.
What will happen instead is an accelerated automation of value-thin, repetitive, and rote portions of white-collar work (Note: this means people will lose their jobs), while massively empowering the workers that orchestrate the automation of the former – the orchestrators. Paradoxically, this might drive additional demand to tackle net-new value-thin tasks that weren’t viable previously due to cost structures, giving even more power to orchestrators.
We see this underway in the market today with (1) the erosion of entry-level jobs, typically associated with value-thin labor, now being automated (or outsourced): data entry, memo drafting, portions of software engineering, and others; and (2) the explosion in vibe-coded software.
Chris Loy alludes to both of these dynamics in his essay on the “Industrialization of Software”:
Industrialisation of production, in any field, seeks to address both of these limitations at once, by using automation of processes to reduce the reliance on human labour, both lowering costs and also allowing greater scale and elasticity of production. Such changes relegate the human role to oversight, quality control, and optimisation of the industrial process.
The first-order effect of this change is a disruption in the supply chain of high-quality, working products. Labour is disintermediated, barriers to entry are lowered, competition rises, and the rate of change accelerates. All of these effects are starting to be in evidence today, with the traditional software industry grappling with the ramifications.
A second-order effect of such industrialisation is to enable additional ways to produce low-quality, low-cost products at high scale.
I would argue that orchestrators play an important role as stewards of quality and taste, helping to stem a flood of low-quality software unleashed on users. “Taste” being AI parlance to signify the ability to create deeply differentiated product experience that stems from vision connected to insight[2].
A real-life example
A friend’s fashion brand provides a practical example. Every year, they do photo shoots for their newest products. This implies hiring models, paying for image rights (usually 24–36 months, after which they expire), makeup, lighting, travel expenditures, external photographers, physical spaces, products, pre-production, post-production, and more. This is an intense and expensive logistical effort. In 2025, it turns out that AI can now generate roughly 80% of their image needs using standard inputs (product photos and model photos). You might think this would put photographers and models out of work.
But here is the rub. The photographers that used to do the photo shoots are now orchestrating the generation of images.
Why? Because AI tools, by default, are generic. They need users to teach them how to do something well in a given context. To produce excellent and extraordinary images that fit a brand, you need specificity. In short, you need “taste.” In the context of my friend’s business, the person best positioned to create great brand photos is the photographer. By my friend’s estimate, the company now spends 50% less on shoots than before.
How orchestrators lever taste
You can now see how job roles will transform from “doing the thing” to “orchestrating the things.” An “orchestrator” is a mix of an engineering and managerial role that drives fleets of agents or agent tooling to accomplish the goals set out by a specific company function (marketing, sales, etc.).
Each function can have many orchestrators, but this implies a massive shrinkage in rank-and-file employees fulfilling value-thin work, as this work gets automated and subsumed by fleets of agents.
The key to success in these roles is that they are taste-levered. They reward employees who use their understanding of underlying business value propositions and their connection to users to hypercharge their vision of deliverable work through agents. Agents allow orchestrators to lever their taste.
Not everyone will be cut out for orchestrator roles. Low-agency employees who “defaulted” into white-collar jobs will be severely punished as high-agency orchestrators automate their work.
At the same time, this shift offers a massive number of opportunities for orchestrators. To become an orchestrator, three things are needed:
- Willingness, curiosity, and drive to engage with new AI tooling. As the field is shifting rapidly, practitioners with a strong mental map of available and emerging tools have a clear advantage.
- Taste, a much-quoted and hard-to-grasp concept. It forms the core value proposition and moat of white-collar work in the future, imbuing fleets of agents with abilities and vantage points that are unique to the operator. Strong convictions about “what feels right” in a given line of work, combined with the ability to iterate rapidly based on user feedback, matter most.
- Technical abilities to use the tools: planning, delegating, training, evaluating, and adjusting agents. Today these abilities are still quite technical, but better tooling over time will lower barriers to entry.
The tool space for orchestrators is still early. There are domain-specific tools that attempt to move in this direction (Clay, Harvey, and others), but they do not provide full automation. Claude Code and Cursor, which are supposedly the most advanced of these, are still fairly technical and not easy to use due to text-based interfaces such as the command line or IDEs[3]. There are also new job titles emerging, such as “GTM engineer,” but I believe “orchestrator” is the more apt term.
What does this mean from a societal standpoint?
A massive compression in white-collar jobs. Far fewer “bullshit jobs,” as these will be performed by AI, including many layers of middle management. This will have downstream consequences.
The value of “soft” degrees like marketing and business will trend toward zero. Prestigious universities, where value is derived from signaling and networking, will likely be exempted. Graduates filling new orchestrator roles will likely come from more technical backgrounds, with degrees focused on AI tooling, and we will need far fewer of them to produce the same output. Entry-level jobs, therefore, are not dead; they are transforming, as much of their former value can now be realized through different means (i.e., AI and software).
I view this as a net positive. We allocate too much capital to “bullshit jobs”—how many management consultants and KPMG accountants do we really need? This frees up labor for other types of occupations that are in short supply (engineers, nurses, caretakers, craftsmen) and breaks the idea that a university degree is a prerequisite for success.
This period of transformation will be painful. As traditional career paths become unviable, several graduating classes will experience difficulty as they attempt to follow those paths without additional preparation. There will not be that many orchestrator jobs. It seems reasonable to expect new graduates seeking these roles to demonstrate advanced AI proficiency through projects they have driven themselves.
University degrees should no longer be a strict requirement for many orchestrator roles, although degrees from elite universities can still serve as a strong signal. That said, I would prefer even elite universities to drop business and marketing degrees and replace them with liberal arts programs or a “studium generale” that teaches the critical thinking skills required for taste-levered work.
Final notes
- 2026 feels like the breakout year for orchestrators. They will not take over the job market entirely, but job openings focused on this role will become increasingly common. Developing tools for orchestrators presents a large market opportunity, alongside new monetization challenges.
- Matthias rightly pointed out that some jobs, even if value-thin, will remain hard to automate. Examples include roles with regulatory requirements (notaries in Germany, for example) or work that is not yet digitized, although the industrialization of software may help here.
- On the flip side, the acceleration may arrive much faster than previously imaginable, with 95%+ of white-collar jobs potentially falling into the value-thin category sooner than expected.
As a semi-technical person with broad domain knowledge and strong opinions, I am personally very excited about this emerging world.
Thank you to Jimena Mondragon, Matthias Plappert, Clemens Viernickel, Fabian Ruffy, James Moran, Mengfei Gao, and Johannes Theobald for their in-depth feedback on this post.
When I refer to AI throughout this essay, I mostly mean LLMs, including multimodal and media generators. ↩︎
Some good pointers here: https://www.workingtheorys.com/p/taste-is-eating-silicon-valley. ↩︎
Integrated development environments, where most software engineers do most of their work. ↩︎