Marc Andreessen has a new thesis about talent in the AI era: go wide, not deep. Speaking at the Andreessen Horowitz LP Conference in Las Vegas in May, the venture capitalist argued that founders who can claim competence across six to eight disciplines will outperform the specialists who dominated the last few decades of tech.
"I think there's basically two ways to really have a differentiated edge in general: go deeper or go broad," Andreessen said on the TBPN podcast. And for most fields, he's betting on broad.
The logic is seductive. If AI can dive deep into any topic on demand, the human advantage shifts to synthesis and integration. Specialists become less essential when a large language model can approximate their expertise in seconds. The generalist, meanwhile, becomes the conductor orchestrating an AI-powered symphony.
The exceptions tell the story
But listen closely and the caveats start to accumulate. Andreessen explicitly exempts biotech and foundation model research from his generalist gospel. "There are domains in which that really matters," he said. "The deeper you are, the better."
This is where the thesis gets interesting, and not necessarily in ways Andreessen intends. The domains he exempts from the generalist rule happen to align with where a16z deploys its largest checks. The firm participated in Databricks' $10 billion round, backs foundation model companies like Mistral, and has invested heavily in biotech through its dedicated bio fund.
So the message to most founders is: be a generalist and use AI as your depth amplifier. But if you're building the actual AI infrastructure or doing serious biotech R&D, you'd better be a world-class specialist. The convenient alignment between Andreessen's advice and his investment strategy doesn't invalidate the thesis, but it warrants some skepticism about its universal applicability.
What "six to eight things" actually means
Andreessen's claim that the best future founders will be "quite skilled at six or eight things" sounds precise but resists easy interpretation. Skilled at what level? Product sense, sales, technical architecture, hiring, finance, marketing, operations, customer development, all at once?
The framing echoes a concept that's circulated in management circles for years: the T-shaped professional, someone with deep expertise in one domain and functional literacy across adjacent ones. Carol Dweck's growth mindset research, which Satya Nadella famously adapted into Microsoft's "learn-it-all" culture transformation, makes a similar point about cognitive flexibility.
But there's a meaningful difference between "learn-it-all" and "know-it-all across six domains." Nadella's formulation emphasizes continuous learning and intellectual humility. Andreessen's version implies a broader baseline competence that few people actually possess.
The founders who fit this description already run the companies VCs fight to fund. Jeff Bezos, Elon Musk, the usual suspects. Whether AI tools can manufacture this kind of polymathic capability in people who don't already have it remains an open question that Andreessen's thesis conveniently elides.
The talent war suggests otherwise
Other voices in the AI industry tell a different story. Demis Hassabis at Google DeepMind has highlighted the ongoing scramble for elite AI researchers. Meta's aggressive talent acquisition, which Hassabis has called a "rational" response to being behind in the foundation model race, reflects a market that still places enormous premiums on specialized depth.
The divergence makes sense when you separate the layers of the AI stack. At the infrastructure layer, building the models themselves, deep specialization remains non-negotiable. The research scientists training frontier models need years of focused expertise in architecture, optimization, and scaling laws.
But at the application layer, where most startups operate, Andreessen's thesis has more purchase. If you're building an AI-powered legal tool or an automated sales assistant, you probably don't need to understand transformer architectures at the attention-head level. You need to understand law or sales at the practitioner level, then wield AI tools to amplify that knowledge.
The question is which layer grows faster in the coming decade. A16z is betting on applications, which fits the generalist thesis. But if the foundation model layer expands and absorbs more of the value chain, the specialists might have the last laugh.
The investment math
Andreessen Horowitz has deployed substantial capital into AI in recent years. The firm is reportedly raising a $20 billion fund focused on growth-stage AI companies, which would exceed all US venture funding from Q1 2025 combined. Its portfolio includes OpenAI, Databricks, and a growing list of AI application startups.
This level of commitment to AI makes Andreessen's pronouncements about the field something other than disinterested analysis. He's talking his book, which is fine as long as everyone recognizes it. The generalist thesis supports a world where AI applications proliferate rapidly, driving demand for product-minded founders who can integrate tools across domains rather than build the underlying technology.
That world might materialize. The alternative, where AI development concentrates in a handful of foundation model labs that capture most of the value, would favor the specialists Andreessen says matter less in "most fields." His confidence in the former scenario is easier to understand when you look at where a16z's money sits.
The FTC's recent scrutiny of major AI investments adds another wrinkle. If regulatory pressure limits the ability of foundation model companies to consolidate the market, the application layer where generalists thrive becomes more attractive. Andreessen's political positioning, including his advisory role in the Trump administration, might factor into this calculus.
None of this makes his thesis wrong. But the founder evaluating career strategy should probably weight it accordingly.
