DOIST'S "LESS AI" STRATEGY FOR TODOIST: RELIABILITY OVER HYPE
Doist argues that shipping fewer AI automations in Todoist leads to simpler, more reliable systems. In a [The New Stack piece](https://thenewstack.io/doist-ai-...
Doist argues that shipping fewer AI automations in Todoist leads to simpler, more reliable systems.
In a The New Stack piece, Doist outlines why they’re resisting the urge to wire AI into every workflow in Todoist. The focus: keep core behavior predictable, reduce failure modes, and avoid opaque, hard-to-debug paths.
The takeaway for teams: treat LLMs as optional enhancers around well-instrumented, deterministic flows rather than the backbone. That bias keeps SLAs stable, costs sane, and rollbacks straightforward.
Reliability and latency targets are easier to hit with deterministic flows than with probabilistic AI steps.
Simpler, non-AI paths cost less to run and are faster to debug, which reduces on-call load.
-
terminal
A/B core automations: rules-first vs LLM-assisted; measure task accuracy, latency p95, and incident count.
-
terminal
Introduce kill switches and fallbacks for any AI step; verify graceful degradation under model timeouts or bad outputs.
Legacy codebase integration strategies...
- 01.
Audit AI-in-the-loop automations and replace brittle steps with rules or heuristics where accuracy/latency slip.
- 02.
Gate remaining AI calls behind feature flags and circuit breakers; add structured logging for prompt/response traces.
Fresh architecture paradigms...
- 01.
Design core flows as deterministic pipelines and layer AI at the edges (classification, summaries) with strict timeouts.
- 02.
Define SLOs and cost budgets up front; require fallbacks for every AI decision point.
Get daily THE-NEW-STACK + SDLC updates.
- Practical tactics you can ship tomorrow
- Tooling, workflows, and architecture notes
- One short email each weekday