When Software Started Writing Software — A Developer’s History of AI
A ~2,500-word retrospective (Adam “the Developer”, DEV Community, 2026-06-22) tracing a 70-year arc of AI through one lens: where the “intelligence” lived at each stage, and which layer of human work it automated. A dev-tech historical narrative — the genre-sibling of tech-adoption-curve-twenty-years, with the AI arc as its subject. Routed to cluster F (and the wiki’s mechanizing-thought meta-thread).
The five eras
- Symbolic (1950s–80s) — hand-coded rules; ELIZA, expert systems. Intelligence lived in the programmer’s explicit logic.
- Statistical ML (1990s–2000s) — pattern recognition, early neural nets; intelligence moves into learned parameters.
- Deep learning (2012–17) — AlexNet, representation learning; the machine learns its own features.
- Transformers & LLMs (2017–22) — the GPT lineage to ChatGPT’s public launch.
- Agentic systems (2023–present) — multi-step planning, tool use, autonomous loops; agents that write, test, and ship code.
The thesis worth keeping
“Each leap automated a layer of mechanism and left the judgment layer exactly where it was” — so human value hasn’t vanished, it moved upstream, and “developers watched their job description quietly rewrite itself” over three years. This is a clean external statement of two threads the wiki already tracks:
- The mechanizing-thought meta-observation (synthesis): the recurring move across memex/llm-wiki (KM), formal methods, and agent-skills is mechanizing successive layers of human thought — this essay generalizes it to “mechanism automates, judgment relocates,” the same shape one rung more abstract.
- The augment→automate axis: the symbolic→agentic progression is the augment→automate spectrum told as history — and it lands on the same conclusion as the adoption sources, that the open frontier is “how autonomous should the maintaining/coding agent be,” not whether the leap happened.
Why it’s here (cluster F + meta-thread)
It pairs with tech-adoption-curve-twenty-years as the cluster’s second dev-tech historical retrospective — where InfoQ places 20 years of tooling on the adoption curve, this places 70 years of AI on a “which layer got automated” curve, both landing agentic systems in the innovator/early-adopter band. The “judgment moved upstream” payoff is the cross-spoke seam to agentic-tooling-wiki (its augment→automate axis and “value moved to everything around the model”) — the runner-up spoke, where the software-writing-software culmination would live if this were a tooling source rather than a history.
Caveat
A single developer’s opinion retrospective (T4): accessible, no citations or original data, and the era boundaries/dates are the author’s tidy narrative, not a sourced timeline. Value is the framing (“automate the mechanism, judgment relocates”), not the history’s precision.
Related
tech-adoption-curve-twenty-years · technology-adoption-curve · tools-for-thought · agent-skills · llm-wiki · crossing-the-chasm