scaffolding
In plain wordsThe code, prompts and tools built around an AI model so it can carry out tasks it could not manage from a bare prompt — like the scaffolding that lets builders reach every part of a building.
The non-model components that structure how a large language modellarge language model, n.: A computer program trained on enormous amounts of text to predict the words that come next. At vast scale, that simple skill lets it chat, answer questions, translate, summarize and write code. is used to perform a task: system prompts and instructions, tooltool, n.: Any outside capability an AI model is allowed to use — a web search, a calculator, a code runner, a database or an app — described to the model so it can ask for it by name. definitions and output parsers, chainsprompt chaining, n.: A way of getting a big job done with a relay of AI requests: each handles one stage and passes its result to the next — pull out the facts, write, polish — with automatic checks between stages. of calls, memory and retrieval, planning or critique steps, and the control logic connecting them. Scaffolding turns a text predictor into a system that executes multi-step work, and better scaffolding can raise measured capability substantially without changing the weights — which is why dangerous-capability evaluationsdangerous capability evaluation, n.: A test of whether an AI system can do things that could cause severe harm in the wrong hands or on its own — such as helping make a bioweapon, hacking computers or copying itself. and agent benchmarksbenchmark, n.: A fixed, shared test — a set of questions or tasks with an agreed way of scoring — used to compare AI systems with one another and to track progress over time. use strong scaffolds so as not to underestimate a model. Usage overlaps with harnessharness, n.: The software around an AI model that gives it instructions and tools, carries out its actions and decides when it stops. The model is the engine; the harness is the rest of the car.: some writers treat the words as synonyms, while others distinguish the harness (the runtime that executes the loop, tools and permissions) from the scaffolding (the prompts, tool descriptions and logic that shape behavior).
With better scaffolding — a planner, a code runner and a retry step — the same model solved twice as many tasks.
Origin
The building-site metaphor became common in AI evaluation and safety writing around 2023 for the wrapper code around models, such as AutoGPTAutoGPT, prop. n.: A 2023 open-source program that gave GPT-4 a goal and let it run on its own — planning, searching the web and writing files in a loop — becoming the first viral 'autonomous AI agent'.; evaluators at METR and government AI safety institutes used 'scaffold' for the agent code under which a model is tested. Education has an older, separate sense: temporary support for a learner (David Wood, Jerome Bruner and Gail Ross, 1976).
Usage note
In software engineering 'scaffolding' also means generated boilerplate for a new project, an unrelated sense. Compare capability elicitationcapability elicitation, n.: The work of drawing out the best an AI can do on a test — with better instructions, tools, retries or extra training — so that its abilities are not underestimated., which deliberately uses strong scaffolding to reveal a model's upper-bound abilities.
Related
- harnessharness, n.: The software around an AI model that gives it instructions and tools, carries out its actions and decides when it stops. The model is the engine; the harness is the rest of the car.
- agent loopagent loop, n.: The repeating cycle at the heart of an AI agent: the model decides what to do next, software carries it out, the result is shown to the model, and the cycle repeats until the task is finished.
- capability elicitationcapability elicitation, n.: The work of drawing out the best an AI can do on a test — with better instructions, tools, retries or extra training — so that its abilities are not underestimated.
- agent frameworkagent framework, n.: A software library that gives developers ready-made parts for building AI agents — model connections, tool handling, memory, multi-agent coordination and logging — so they need not write everything from scratch.
- prompt chainingprompt chaining, n.: A way of getting a big job done with a relay of AI requests: each handles one stage and passes its result to the next — pull out the facts, write, polish — with automatic checks between stages.
- tool usetool use, n.: An AI model's ability to reach beyond its own knowledge by operating outside tools — searching the web, running code, using a calculator or an app — and folding the results into its work.
- modelmodel, n.: A simplified working version of something. In AI it usually means the trained program, a big set of learned numbers, that takes an input and produces an answer, prediction or piece of content.
- fine-tuningfine-tuning, n.: Taking an AI model that has already learned a lot and training it a little more on a smaller, focused set of examples so it gets better at one job or behaves a certain way.
Sources
- Paniego, S. & Roy Gosthipaty, A. (2026). Harness, Scaffold, and the AI Agent Terms Worth Getting Right. Hugging Face blog, 25 May 2026. huggingface.co/blog/agent-glossary
Mentioned in
SWE-benchSWE-bench, prop. n.: A test that asks AI to fix real bugs and add features in real open-source Python projects, checked by running each project's own tests; the best-known yardstick for AI coding agents., agent loopagent loop, n.: The repeating cycle at the heart of an AI agent: the model decides what to do next, software carries it out, the result is shown to the model, and the cycle repeats until the task is finished., capabilitycapability, n.: Something an AI system is able to do, such as writing working code or solving math problems, and how well. Measuring capabilities is how researchers track progress and spot dangerous skills., evaluation harnessevaluation harness, n.: Software that puts AI models through tests in a fixed, repeatable way — feeding in the questions, collecting the answers and scoring them — so that every model is judged identically., harnessharness, n.: The software around an AI model that gives it instructions and tools, carries out its actions and decides when it stops. The model is the engine; the harness is the rest of the car., autonomous replication and adaptationautonomous replication and adaptation, n.: An AI's ability to keep itself running and spreading without human help — earning money, renting computers, copying itself and fixing problems — which safety testers check for., capability elicitationcapability elicitation, n.: The work of drawing out the best an AI can do on a test — with better instructions, tools, retries or extra training — so that its abilities are not underestimated., unhobblingunhobbling, n.: Fixing the simple ways a capable AI model is held back — such as by giving it tools, memory or time to think — so that abilities it already has can actually be used.