Coding is changing before our very eyes, and AI leaders are using some interesting terms to describe what’s coming.
Aravind Srinivas, CEO of Perplexity AI, has a clear framework for how he sees the evolution of AI coding tools — and it’s more radical than most people realise. In a recent conversation with Jason Calacanis, Srinivas laid out three distinct paradigms: autocomplete, auto-diff, and what he calls auto-outcomes. Each one represents a fundamentally different relationship between the developer and the code. The direction of travel, in his view, is toward a world where developers may never read a single line of code at all.
“The framing should be around what does solving (coding) mean,” Srinivas said. “Maybe think of this as paradigms — Cursor, GitHub Copilot — like autocomplete. You’re trying to complete a few lines of code, but you are writing code largely.”
That first paradigm — the era of tools like GitHub Copilot — is already giving way to something different. Srinivas describes the second paradigm as “auto-diff,” referencing tools like Cursor and command-line coding agents that operate at a higher level of abstraction.
“Claude Code, Codex, as command line interfaces — you can almost think of it as auto-diff. You’re looking at the diff: the new lines of code added and the existing lines of code subtracted. You’re not actually autocompleting anymore. You’re operating at a different abstraction of changes.”
But it’s the third paradigm where things get genuinely interesting. Srinivas argues that the next phase won’t be about reading diffs at all.
“The next paradigm is just going to be auto-outcomes. You’re going to look at the outcome. You’re not even going to look at the diff, you’re not going to read any line of code. You’re going to look at the outcome, and then you’re going to ask for changes and you’re going to keep iterating. That’s clearly the next thing.”
He also noted that Elon Musk has been thinking along similar lines — and pushing the idea even further. “Elon talks about it. I think he talked about it in one of the xAI all-hands live streams where he said you’re going to just output the binary — which is a wild thing to think about.”
The shift Srinivas is describing is already underway. The coding tool landscape in 2026 has moved well beyond autocomplete, with Cursor, Claude Code, and GitHub Copilot now operating as agentic systems capable of multi-file editing, autonomous planning, and deep codebase reasoning. The question is no longer whether to use AI in a development workflow — it’s how much of the workflow to hand over entirely.
That question has serious implications for the software industry at large. Srinivas has previously argued that AI will cause job losses in the short term, with fewer people needed to accomplish the same amount of work. He has also predicted that Indian IT companies will hire fewer people in the AI era, as the cost arbitrage advantage of human labor erodes. The auto-outcomes paradigm he describes would accelerate both trends — if developers are evaluating outputs rather than writing or reviewing code, the number of engineers required to ship software could fall dramatically.
Perplexity itself has been moving in this direction. The company recently launched Computer, an agentic tool that lets users delegate research, writing, and coding tasks without deep technical setup. The product is a direct expression of the auto-outcomes philosophy: users define what they want, not how to get there.
The “output the binary” framing that Srinivas attributes to Musk is the logical endpoint of this trajectory — a world where AI systems produce deployable software directly from intent, with no human-readable intermediate step. Whether that’s 2 years away or 10 is an open question. But the direction, as Srinivas sees it, is not.