typing was never the hard part

DevOpsDays Austin 2026

Ian Littman / @ian@phpc.social / @ian.im / @iansltx

Slides at ian.im/doda26

Warning: we're gonna talk about LLMs (#aI)

  1. What I think LLMs are good for
  2. Recent history highlights (since last year's DevOpsDays Austin)
  3. What this seems to mean software dev workflows
  4. What this seems to mean for DevOps workflows

 

I'll include some implementation details on my setup. Find me after to chat more.

 

Slides are mine. LLMs didn't touch them.

ian.im/doda26

What LLMs are good for

  1. Easy to describe acceptance criteria
  2. Toilsome to implement
  3. Following an existing pattern
  4. Straightforward (preferably automated) to verify
  5. Small units of work (this is no longer a constraint)

ian.im/doda26

History: Better models && better harnesses

ian.im/doda26

Late May 2025

  • Claude Sonnet 4 released by Anthropic
  • Claude Code went GA
  • I start using Sonnet ~1mo later, and then more in August
  • Opus existed but was $$$; Sonnet was a situational toil-reduction pick
  • Already easy to find a model that was better at jq/bash/regex than I was

ian.im/doda26

september 2025

  • Claude Sonnet 4.5 released by Anthropic
  • Useful for a larger selection of toil
  • Used occasionally by me
  • Qwen3-Next, GLM 4.6 released

ian.im/doda26

Late November 2025

  • Claude Opus 4.5 released by Anthropic
  • Step change in capabilities
  • Significantly less expensive
  • I started throwing it larger tasks ~year-end
  • GLM 4.7 released in December

ian.im/doda26

February 2026

  • Sonnet 4.6 + Opus 4.6 released
  • GPT 5.3 Codex released
  • Qwen3.5 released (step change on local model ability)
  • GLM 5 released
  • Kimi K2.5 released in January (I used it a bit)

ian.im/doda26

march 2026

  • Claude Code instability
  • Claude plan rate limit revisions
  • Claude code review is more widely available
  • GPT-5.4 released

ian.im/doda26

April 2026

  • Releases (non-exhaustive list)
    • GPT-5.5
    • GLM 5.1
    • Qwen3.6
    • Kimi K2.6
    • DeepSeek V4
    • Opus 4.7
  • Rugpulls
    • GitHub Copilot
    • Claude Enterprise

ian.im/doda26

May 2026

  • Multi-token prediction (faster local models)

ian.im/doda26

Okay, that was a lot

ian.im/doda26

This means that...

ian.im/doda26

  • Using an LLM is increasingly likely to cost real money
  • Business models based on sustained LLM subsidies are time bombs
  • Doing less (token count) with less (cheaper/simpler models) matters
  • If deterministic code and LLMs can both do a thing comparably well,
    pick (or generate) the former

Software dev

ian.im/doda26

  • Using an LLM is increasingly likely to cost real money
  • Business models based on sustained LLM subsidies are time bombs
  • Doing less (token count) with less (cheaper/simpler models) matters
  • If deterministic code and LLMs can both do a thing,
    pick (or generate) the former

let's talk about subsidies

ian.im/doda26

  • LLM lab plans (OpenAI, Anthropic)
  • GitHub Copilot
  • OpenCode Go (less egregious?)
  • Compared to paying APIs per-token

Thanks!

Questions? Find me here / @ian@phpc.social / @ian.im / @iansltx

Slides: https://ian.im/doda26