What Can We Learn from the Internet "Revolution" for AI? (intentful.ueberproduct.de)

The author looks back on the messy, contradictory early internet (circa 1995) to draw lessons for today's AI boom: infrastructure revolutions initially feel chaotic, full of half-baked ideas, failed paradigms and unexpected winners. Early web work involved many disjointed protocols (finger, NNTP, WAIS, Archie, telnet) until HTTP/unified browsers made value obvious; compute limits then constrained promising demos (early Siri-like systems), and many “smart” ideas (mobile-agent platforms) died despite solid theory. Practical breakthroughs often came from serendipity—language-detection Markov chains and a crawler for Chinese pages evolved into search startups (FLP → fireball) and commercial products—while simple consumer-facing services (classifieds, payments, voice/communications) eventually became huge despite seeming mundane at first. For AI/ML practitioners the message is concrete: don’t conflate infrastructure with value — expect long, nonlinear timelines, weird failures, and role evolution (product management, agile practices) as core parts of maturing ecosystems. Technical implications include the importance of system-level constraints (compute, caching, indexing), language and data engineering, and user-centered product discovery over purely academic demos. History suggests the unexpected, incremental applications and better product–customer loops—not only core models or flashy demos—will create the durable value in AI.