Software's Changing Cost Equation
Cost of Tokens vs. Engineers
Two to three years ago the Software Cost Equation was primarily the cost of developers / engineers / architects plus the cost of computing infrastructure (e.g. “compute”). That math is changing as LLMs increasingly automate the Software Engineering function.
Now, especially for small companies like ours that can move fast, LLMs are replacing Software Engineering talent, or at a minimum reducing demand. When we start ramping up I want to scale a Software organization where each person is well-versed in leveraging LLMs given that the models are a force multiplier for software engineers (SWEs).
Anthropic’s founder Dario Amodei swears that in 1-2 years Anthropic’s LLMs will automate all SWE job functions, tasks, everything. Maybe, although there is no denying that directionally Dario is correct. More and more of the SWE function will be automated.
Therefore, the new Software cost equation becomes:
the cost of compute plus,
the cost of tokens (which sit above the chip layer, but I think of tokens as compute cost as well) plus,
the cost of developers / engineers / architects.
The mix shift (tokens vs. SWEs), will change over time on a per company basis. The cost of tokens at the unit level will come down, but token consumption will move higher as more companies automate the Software Engineering function.