The AI Engineers Building the Future of Silicon Valley

A group of people with handwritten signs gathered on the sidewalk outside Anthropic’s San Francisco office late on a Tuesday afternoon in late March. Some read, “Stop the AI Race,” “Don’t Build Skynet.” The demonstrators were sincere, well-organized, and, depending on who you asked inside the building they were protesting outside of, either genuinely alarmed or fundamentally misinterpreted what was going on on the floors above them. Maybe both. Hanging over a city block in downtown San Francisco, that ambiguity seems to sum up Silicon Valley’s current situation rather well.

Within those offices, as well as numerous others dispersed throughout the Bay Area, a particular type of engineer has emerged as the most in-demand specialist in the technology sector. They are AI engineers, not the traditional software developers who have spent the last 20 years creating applications for human users, nor the research scientists who write papers on model architecture.

They occupy a space in between, creating systems that are increasingly capable of building themselves, setting up agents to carry out tasks that previously required teams on their own, and posing questions about their own professional relevance that most industries don’t face until much later in a technological cycle.

Anthropic has stated openly that Claude, the company’s AI model, now writes about 90% of the code. OpenAI has set an internal goal to deploy what it refers to as a “intern-level AI research assistant” within months after releasing a system that it claimed was crucial to its own creation. These projections are not far away. These are current events taking place in actual buildings with actual addresses, involving actual engineers who stroll past coffee shops and demonstrators on their way to work, as well as the general cacophony of a city that has seen more technological disruptions than any other location on the planet and still manages to appear a little taken aback every time.

Observing all of this, it seems like the most in-demand engineers are being hired to solve a problem that they are also accelerating. At businesses vying for a talent pool that is currently understaffed, the salary figures are astounding—six figures well into the two and three hundreds. Because the alternative is falling behind in a cycle that seems to be compressing timelines in ways that make standard hiring logic seem quaint, startups are paying amounts that would have seemed excessive even during the 2021 funding frenzy. According to a Fortune article, businesses developing AI systems are desperate to hire humans to do the work, despite Gen Z workers’ concerns that AI will eliminate jobs.

“We are starting to see AI progress feed back on itself,” said Nick Bostrom, a Swedish philosopher whose research on AI risk has influenced over ten years of discussion in this field. A moment should be given to that sentence. It’s not outlining a scenario for the future. It describes an already-observable loop in which AI systems are used to create better AI systems, reducing the time between innovations and raising concerns about the true endpoint of that compression. Depending on where you stand in the industry and how much of your professional identity is linked to being the one who does the building, you may find that exciting or unsettling.

This is driven by a straightforward business logic. In a single quarter, the combined quarterly capital expenditures of Amazon, Microsoft, Google, and Meta have come close to $100 billion; funds are pouring into infrastructure, chips, and data centers at a rate that makes earlier technological advancements seem modest. Silicon Valley is no longer primarily known for producing innovative software at low cost and with rapid scalability.

It’s evolving into an infrastructure sector that is more like the steel and railroad firms of the late 19th century than the agile startups of the 2010s. The engineers employed in this shift are more like the civil engineers of a previous industrial moment than the “move fast and break things” coders of the Zuckerberg era; they are creating long-lasting structures in environments that change more quickly than the blueprints can be updated.

It’s worth considering the comparison to more established enterprise software firms. Instead of using elegant simplicity to establish their dominance, Salesforce, ServiceNow, and Workday did so by integrating deeply and messily into the systems that their clients already used. Because removing them would require removing something load-bearing, they became difficult to remove precisely.

Today’s AI companies appear to have an innate understanding of this; their products are more embedded infrastructure than consumer apps, necessitating implementation specialists and services teams as well as the kind of patient, hands-on work that doesn’t look good but gives them a significant competitive advantage. It’s still unclear if these startups will create deep enough moats to matter or if their functions will be absorbed by the major platforms before the moats have a chance to solidify.

Product engineer is a new title that has begun to show up in internal organizational charts and job listings. Because AI tools have increased the productivity of individual engineers to the point where the ratio of engineers to product managers no longer makes organizational sense, this hybrid role, which sits between the engineering function and product management, has emerged.

It’s good news that one person can now perform tasks that previously required multiple people, but it also means fewer people are being hired to fill those multiple roles. The industry is keeping an eye on this math in real time, but the results are still not entirely clear. Even though no one has decided on a term yet, it’s obvious that the engineers still in the room—those creating prompts, setting up agents, and creating systems that create other systems—are doing something truly novel.