Developer argues coding is now a parallel problem: steer many agents, let cheap models implement

Aryvyo · x · 2026-09-19

The author argues that coding tasks today are a highly parallel problem: you can explore the solution space far more efficiently by steering multiple agents at once. In that workflow, a model burning 10x the thinking tokens doesn't appeal to him.

His trade-off: use a cheap model like DeepSeek as the implementation workhorse, since implementation isn't the bottleneck — the real value lies in parallel exploration of the solution space rather than one model digging deep.

Related event: Developer advocates parallel multi-agent coding with cheap models(2 posts)→

Original post →

More from coding & agent

coding & agent channel →