The vocabulary of AI-era technical hiring
Definitions for the vocabulary of AI-era technical hiring: process telemetry, agentic coding assessment, orchestration skill, AI cheating detection, and more.
Agentic coding assessment
An agentic coding assessment is a technical evaluation where the candidate works on a real coding task with their own AI agent — Claude Code or Codex — and the platform captures the full session as process telemetry so reviewers can grade how they orchestrated the AI, not just whether the final code passed tests.
Also known asagentic coding interviewAI agent coding interviewClaude Code assessmentagentic developer evaluationAI-collaboration interviewRead definition →AI cheating detection
AI cheating detection in coding interviews is the practice of identifying whether a candidate is using AI assistance against the assessment's rules — or, in the post-2024 model, whether the AI use the candidate disclosed actually matches what they did. The honest version uses process telemetry to find contradictions; the legacy version uses typing-speed and paste-frequency heuristics that no longer separate cheaters from serious engineers.
Also known asAI cheating detection in interviewsAI cheating preventionAI-proof coding interviewChatGPT cheating detectionAI plagiarism detection in technical hiringRead definition →AI Coding Fluency
AI coding fluency is a developer's skill at getting durable leverage from AI coding agents: right-sizing the context they hand a model, choosing the right model for the task, steering a session instead of re-prompting, and knowing when AI is the wrong tool. It's a learnable skill, not a property of the tool — and because fluent workflows waste fewer tokens, fluency is where lasting cost savings come from. It's coached from a private replay of the developer's own sessions.
Also known asAI fluencyAI-assisted coding skillagent fluencyAI enablementRead definition →AI Spend Attribution
AI spend attribution is the practice of tracing LLM token cost back to the developer, team, repo, and workflow that produced it — turning one opaque provider invoice into per-person, per-workflow accountability. Provider dashboards attribute to the API key and workspace, which only equals a human when every developer has a distinct key and nobody shares a gateway. Session-level attribution reads the developer and workflow from the coding session directly, so it survives shared keys.
Also known asLLM cost attributiontoken spend attributionAI cost allocationper-developer AI costRead definition →AI-enablement platform
An AI-enablement platform is software that helps an engineering organization get more leverage from the AI coding tools it has already adopted — Copilot, Cursor, Claude Code, Codex — by measuring how engineers actually work with those tools and turning that into guidance for leaders and developmental feedback for individual engineers. It sits above the coding tools rather than replacing them.
Also known asAI enablement softwareAI adoption platform for engineeringAI fluency platformdeveloper AI-effectiveness platformRead definition →AI-era technical hiring
AI-era technical hiring is the practice of assessing software engineers under the assumption that they use AI coding agents — Claude Code, Codex — to do their real work. The interview loop measures how the candidate orchestrates the agent, not whether they can write code without one, because the latter no longer predicts on-the-job performance.
Also known asAI-native technical hiringagentic-age technical hiringpost-LeetCode hiringAI-proof technical interviewAI-collaboration interviewRead definition →Context compaction
Context compaction is what an AI coding agent does when a conversation outgrows its context window: it summarizes the history down to a smaller representation and continues from that summary rather than the full transcript. It is a recovery mechanism, not a feature you want to trigger often: compacting costs tokens to perform, discards detail the model may need later, and is usually the visible symptom of a window that was carrying more than the task required.
Also known ascompactionauto-compactcontext summarizationwindow compactionRead definition →Context Efficiency
Context efficiency is the practice of handing an AI coding agent only the tokens a task actually needs, and no more. Because a model pays for every token in its context window on every turn, a bloated window — stale files, whole dependencies pulled in for one function, history that stopped mattering — is repeated waste. Context efficiency is a core habit of AI coding fluency and the direct antidote to the re-read-loop and bloated-context shapes of recoverable token waste.
Also known ascontext window efficiencycontext managementlean contextcontext hygieneRead definition →Model right-sizing
Model right-sizing is the practice of matching the model you use to what the task actually requires, rather than defaulting to the largest one available. Because top-tier models cost several times what smaller ones do per token, routing routine work to a frontier model is one of the easiest ways to spend a rate-limit window without getting anything for it. Right-sizing is not about always choosing the cheap model. It is about the choice being deliberate rather than default.
Also known asright-sizingmodel selectionmodel tieringover-modellingRead definition →Orchestration skill
Orchestration skill is the ability of a software engineer to direct an AI coding agent — like Claude Code — through real engineering work: scoping the problem before prompting, articulating tradeoffs, pushing back when the model is wrong, sequencing tool calls in the right order, and knowing when not to hand work off. It is the skill that replaced raw coding throughput as the primary signal for senior engineers in 2026.
Also known asAI orchestrationAI orchestration skillAI tool orchestrationagent orchestrationdeveloper orchestrationRead definition →Process telemetry
Process telemetry, in technical hiring, is a structured record of how a candidate worked with their AI assistant during a coding session: every prompt they sent, every diff the model produced, every command they ran, and every decision they made along the way — captured as typed events and replayed as a searchable timeline.
Also known assession telemetrycandidate process telemetryworkflow telemetryagentic process captureRead definition →Rate-limit window
A rate-limit window is the rolling period an AI coding subscription meters your usage against, after which your allowance refills. Claude Code runs two at once: a short window measured in hours and a long one measured in days. Hitting the short one costs you a coffee break; hitting the long one can cost you the rest of your week. What makes the long window hard to manage is that it is spent gradually and invisibly, so by the time you notice you are near it, the habits that spent it are days behind you.
Also known asusage windowrolling limitweekly limitsession limitRead definition →Recoverable Token Waste
Recoverable token waste is the portion of AI coding spend that produced no output worth its cost and can be cut without cutting work — as opposed to leverage, the spend that actually shipped something. It has nameable shapes: re-read loops over the same files, context windows bloated with irrelevant tokens, redundant runs of a workflow that already succeeded, and an oversized model routed to a trivial task. Naming and cutting it is the optimize stage of TokenOps.
Also known astoken wastewasted AI spendrecoverable AI spendLLM cost wasteRead definition →TokenOps
TokenOps is the discipline of observing, attributing, and optimizing LLM token spend across an engineering team. It is to AI-agent token spend what FinOps is to cloud spend and what MLOps is to model deployment: a practice that turns one opaque provider invoice into per-developer, per-repo, per-workflow accountability, separates recoverable waste from real leverage, and closes the loop with budgets and trends.
Also known asFinOps for AILLM token spend managementAI coding cost managementtoken spend optimizationAI agent cost attributionRead definition →
Read the process,
not just the commit.
Twelve founding teams will ship this with us. A technical screen that can't tell paste from craft isn't neutral. It's a ~$200K coin-flip you won't catch for months. If you hire 5+ engineers a year, we should talk.