CtrlK
BlogDocsLog inGet started
Tessl Logo
PluginPublic

ainativedev/aidevcon-2026-nyc

AI Native DevCon 2026 New York — practical agent skills grounded in conference talks

Quality

86%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Low

Low-risk findings worth noting

Overview
Quality
Evals
Security
Files

application-template.mdskills/talk-singhal-operating-agents-at-scale/references/

Coding-agent configuration change brief

This is a Tessl adaptation of Singhal’s operating approach. Use it for one decision about a specific setup. Fill missing evidence with “unknown”; do not replace it with an assumed improvement.

Candidate and baseline

  • Decision to make:
  • Current version and candidate version:
  • Model, harness, reasoning, context policy, plugins, and other changed settings:
  • Components changing together:
  • Affected agent-use layer, cohort, and representative tasks:
  • Expected benefit and plausible quality, caching, or reliability tradeoff:

Measurement contract

Define adoption as access, setup, recent use, and retained use with explicit denominators and dates. Define delivery outcomes separately: merged work, follow-up changes, reverts, and cost per merged change. Choose an observation window suitable for those delayed outcomes.

List tokens, sessions, requests, tool calls, and cache behavior only as diagnostic signals. Explain how rising activity could indicate either useful work or retries.

Evidence checkpoints

CheckpointRequired evidenceObserved resultRemaining blind spot
GatewayVersion-tagged cost, latency, errors, identityAgent behavior
TracesCommon event structure, timings, retriesAlternative setup performance
Internal evaluationsPinned code, real tasks, repeated runs, graded explanationsUnrepresented real work
Small rolloutBaseline comparison, cohort, observation periodDelayed outcomes

Record any additional data collection separately from the structure-and-timing approach described in the talk. Do not assume prompt or code collection is necessary.

Failure review

For each task, inspect tests, intended patterns in the diff, agent actions, and judge explanation. Record individual regressions and repeated-run variation even when average quality is unchanged. Examine cache reuse and whether the candidate overrides existing model choices.

Decision and recovery

State retain, investigate, trial, expand, or revert; cite the supporting evidence and what is still pending. Identify the existing owner for configuration, approved models, vetted tools, and merge controls. Specify the prior version, rollback trigger, observation owner, and next review date. Distinguish this proposed decision from any action actually taken.

skills

talk-singhal-operating-agents-at-scale

tile.json