The best way to Write a Good Spec for AI Brokers – O’Reilly

This put up first appeared on Addy Osmani’s Elevate Substack e-newsletter and is being republished right here with the writer’s permission.

TL;DR: Purpose for a transparent spec masking simply sufficient nuance (this will likely embrace construction, fashion, testing, boundaries. . .) to information the AI with out overwhelming it. Break massive duties into smaller ones versus holding all the things in a single massive immediate. Plan first in read-only mode, then execute and iterate constantly.

“I’ve heard loads about writing good specs for AI brokers, however haven’t discovered a strong framework but. I might write a spec that rivals an RFC, however sooner or later the context is simply too massive and the mannequin breaks down.”

Many builders share this frustration. Merely throwing a large spec at an AI agent doesn’t work—context window limits and the mannequin’s “consideration price range” get in the best way. The secret is to put in writing sensible specs: paperwork that information the agent clearly, keep inside sensible context sizes, and evolve with the mission. This information distills finest practices from my use of coding brokers together with Claude Code and Gemini CLI right into a framework for spec-writing that retains your AI brokers centered and productive.

We’ll cowl 5 rules for nice AI agent specs, every beginning with a bolded takeaway.

1. Begin with a Excessive-Degree Imaginative and prescient and Let the AI Draft the Particulars

Kick off your mission with a concise high-level spec, then have the AI increase it into an in depth plan.

As a substitute of overengineering upfront, start with a transparent aim assertion and some core necessities. Deal with this as a “product temporary” and let the agent generate a extra elaborate spec from it. This leverages the AI’s energy in elaboration when you preserve management of the path. This works nicely except you already really feel you might have very particular technical necessities that have to be met from the beginning.

Why this works: LLM-based brokers excel at fleshing out particulars when given a strong high-level directive, however they want a transparent mission to keep away from drifting astray. By offering a brief define or goal description and asking the AI to provide a full specification (e.g., a spec.md), you create a persistent reference for the agent. Planning upfront issues much more with an agent: You’ll be able to iterate on the plan first, then hand it off to the agent to put in writing the code. The spec turns into the primary artifact you and the AI construct collectively.

Sensible strategy: Begin a brand new coding session by prompting 

You're an AI software program engineer. Draft an in depth specification for
[project X] masking aims, options, constraints, and a step-by-step plan.

Preserve your preliminary immediate high-level: e.g., “Construct an online app the place customers can
observe duties (to-do record), with consumer accounts, a database, and a easy UI.”

The agent would possibly reply with a structured draft spec: an summary, function record, tech stack ideas, knowledge mannequin, and so forth. This spec then turns into the “supply of reality” that each you and the agent can refer again to. GitHub’s AI workforce promotes spec-driven improvement the place “specs grow to be the shared supply of reality…residing, executable artifacts that evolve with the mission.” Earlier than writing any code, assessment and refine the AI’s spec. Make sure that it aligns along with your imaginative and prescient and proper any hallucinations or off-target particulars.

Use Plan Mode to implement planning-first: Instruments like Claude Code supply a Plan Mode that restricts the agent to read-only operations—it may possibly analyze your codebase and create detailed plans however gained’t write any code till you’re prepared. That is splendid for the planning part: Begin in Plan Mode (Shift+Tab in Claude Code), describe what you need to construct, and let the agent draft a spec whereas exploring your present code. Ask it to make clear ambiguities by questioning you concerning the plan. Have it assessment the plan for structure, finest practices, safety dangers, and testing technique. The aim is to refine the plan till there’s no room for misinterpretation. Solely then do you exit Plan Mode and let the agent execute. This workflow prevents the widespread entice of leaping straight into code technology earlier than the spec is strong.

Use the spec as context: As soon as accredited, save this spec (e.g., as SPEC.md) and feed related sections into the agent as wanted. Many builders utilizing a robust mannequin do precisely this. The spec file persists between classes, anchoring the AI each time work resumes on the mission. This mitigates the forgetfulness that may occur when the dialog historical past will get too lengthy or when it’s important to restart an agent. It’s akin to how one would use a product necessities doc (PRD) in a workforce: a reference that everybody (human or AI) can seek the advice of to remain on observe. Skilled of us usually “write good documentation first and the mannequin might be able to construct the matching implementation from that enter alone” as one engineer noticed. The spec is that documentation.

Preserve it aim oriented: A high-level spec for an AI agent ought to deal with what and why greater than the nitty-gritty how (a minimum of initially). Consider it just like the consumer story and acceptance standards: Who’s the consumer? What do they want? What does success seem like? (For instance, “Consumer can add, edit, full duties; knowledge is saved persistently; the app is responsive and safe.”) This retains the AI’s detailed spec grounded in consumer wants and consequence, not simply technical to-dos. Because the GitHub Spec Equipment docs put it, present a high-level description of what you’re constructing and why, and let the coding agent generate an in depth specification specializing in consumer expertise and success standards. Beginning with this big-picture imaginative and prescient prevents the agent from dropping sight of the forest for the bushes when it later will get into coding.

2. Construction the Spec Like a Skilled PRD (or SRS)

Deal with your AI spec as a structured doc (PRD) with clear sections, not a unfastened pile of notes.

Many builders deal with specs for brokers very similar to conventional product requirement paperwork (PRDs) or system design docs: complete, well-organized, and straightforward for a “literal-minded” AI to parse. This formal strategy provides the agent a blueprint to comply with and reduces ambiguity.

The six core areas

GitHub’s evaluation of over 2,500 agent configuration information revealed a transparent sample: The simplest specs cowl six areas. Use this as a guidelines for completeness:

1. Instructions: Put executable instructions early—not simply instrument names however full instructions with flags: npm check, pytest -v, npm run construct. The agent will reference these always.
2. Testing: The best way to run assessments, what framework you utilize, the place check information reside, and what protection expectations exist.
3. Mission construction: The place supply code lives, the place assessments go, the place docs belong. Be specific: “src/ for utility code, assessments/ for unit assessments, docs/ for documentation.”
4. Code fashion: One actual code snippet displaying your fashion beats three paragraphs describing it. Embrace naming conventions, formatting guidelines, and examples of excellent output.
5. Git workflow: Department naming, commit message format, PR necessities. The agent can comply with these should you spell them out.
6. Boundaries: What the agent ought to by no means contact—secrets and techniques, vendor directories, manufacturing configs, particular folders. “By no means commit secrets and techniques” was the only commonest useful constraint within the GitHub examine.

Be particular about your stack: Say “React 18 with TypeScript, Vite, and Tailwind CSS,” not “React mission.” Embrace variations and key dependencies. Imprecise specs produce imprecise code.

Use a constant format: Readability is king. Many devs use Markdown headings and even XML-like tags within the spec to delineate sections as a result of AI fashions deal with well-structured textual content higher than free-form prose. For instance, you would possibly construction the spec as:

# Mission Spec: My workforce's duties app


## Goal
- Construct an online app for small groups to handle duties...


## Tech Stack
- React 18+, TypeScript, Vite, Tailwind CSS
- Node.js/Specific backend, PostgreSQL, Prisma ORM


## Instructions
- Construct: `npm run construct` (compiles TypeScript, outputs to dist/)
- Check: `npm check` (runs Jest, should cross earlier than commits)
- Lint: `npm run lint --fix` (auto-fixes ESLint errors)


## Mission Construction
- `src/` – Software supply code
- `assessments/` – Unit and integration assessments
- `docs/` – Documentation


## Boundaries
- ✅ All the time: Run assessments earlier than commits, comply with naming conventions
- ⚠️ Ask first: Database schema adjustments, including dependencies
- 🚫 By no means: Commit secrets and techniques, edit node_modules/, modify CI config

This stage of group not solely helps you suppose clearly but in addition helps the AI discover info. Anthropic engineers advocate organizing prompts into distinct sections (like , , , and so on.) for precisely this cause: It provides the mannequin sturdy cues about which information is which. And bear in mind, “minimal doesn’t essentially imply brief”—don’t shrink back from element within the spec if it issues, however preserve it centered.

Combine specs into your toolchain: Deal with specs as “executable artifacts” tied to model management and CI/CD. The GitHub Spec Equipment makes use of a four-phase gated workflow that makes your specification the middle of your engineering course of. As a substitute of writing a spec and setting it apart, the spec drives the implementation, checklists, and process breakdowns. Your major function is to steer; the coding agent does the majority of the writing. Every part has a particular job, and also you don’t transfer to the subsequent one till the present process is absolutely validated:

1. Specify: You present a high-level description of what you’re constructing and why, and the coding agent generates an in depth specification. This isn’t about technical stacks or app design—it’s about consumer journeys, experiences, and what success seems like. Who will use this? What downside does it remedy? How will they work together with it? Consider it as mapping the consumer expertise you need to create, and letting the coding agent flesh out the small print. This turns into a residing artifact that evolves as you study extra.

2. Plan: Now you get technical. You present your required stack, structure, and constraints, and the coding agent generates a complete technical plan. If your organization standardizes on sure applied sciences, that is the place you say so. Should you’re integrating with legacy techniques or have compliance necessities, all of that goes right here. You’ll be able to ask for a number of plan variations to check approaches. Should you make inside docs obtainable, the agent can combine your architectural patterns instantly into the plan.

3. Duties: The coding agent takes the spec and plan and breaks them into precise work—small, reviewable chunks that every remedy a particular piece of the puzzle. Every process needs to be one thing you’ll be able to implement and check in isolation, nearly like test-driven improvement to your AI agent. As a substitute of “construct authentication,” you get concrete duties like “create a consumer registration endpoint that validates e-mail format.”

4. Implement: Your coding agent tackles duties one after the other (or in parallel). As a substitute of reviewing thousand-line code dumps, you assessment centered adjustments that remedy particular issues. The agent is aware of what to construct (specification), how you can construct it (plan), and what to work on (process). Crucially, your function is to confirm at every part: Does the spec seize what you need? Does the plan account for constraints? Are there edge instances the AI missed? The method builds in checkpoints so that you can critique, spot gaps, and course-correct earlier than transferring ahead.

This gated workflow prevents what Willison calls “home of playing cards code”: fragile AI outputs that collapse underneath scrutiny. Anthropic’s Abilities system provides an identical sample, letting you outline reusable Markdown-based behaviors that brokers invoke. By embedding your spec in these workflows, you make sure the agent can’t proceed till the spec is validated, and adjustments propagate routinely to process breakdowns and assessments.

Take into account brokers.md for specialised personas: For instruments like GitHub Copilot, you’ll be able to create brokers.md information that outline specialised agent personas—a @docs-agent for technical writing, a @test-agent for QA, a @security-agent for code assessment. Every file acts as a centered spec for that persona’s conduct, instructions, and limits. That is notably helpful if you need completely different brokers for various duties moderately than one general-purpose assistant.

Design for agent expertise (AX): Simply as we design APIs for developer expertise (DX), take into account designing specs for “agent expertise.” This implies clear, parseable codecs: OpenAPI schemas for any APIs the agent will devour, llms.txt information that summarize documentation for LLM consumption, and specific sort definitions. The Agentic AI Basis (AAIF) is standardizing protocols like MCP (Mannequin Context Protocol) for instrument integration. Specs that comply with these patterns are simpler for brokers to devour and act on reliably.

PRD versus SRS mindset: It helps to borrow from established documentation practices. For AI agent specs, you’ll usually mix these into one doc (as illustrated above), however masking each angles serves you nicely. Writing it like a PRD ensures you embrace user-centric context (“the why behind every function”) so the AI doesn’t optimize for the flawed factor. Increasing it like an SRS ensures you nail down the specifics the AI might want to really generate appropriate code (like what database or API to make use of). Builders have discovered that this further upfront effort pays off by drastically decreasing miscommunications with the agent later.

Make the spec a “residing doc”: Don’t write it and overlook it. Replace the spec as you and the agent make selections or uncover new information. If the AI needed to change the information mannequin otherwise you determined to chop a function, replicate that within the spec so it stays the bottom reality. Consider it as version-controlled documentation. In spec-driven workflows, the spec drives implementation, assessments, and process breakdowns, and also you don’t transfer to coding till the spec is validated. This behavior retains the mission coherent, particularly should you or the agent step away and are available again later. Bear in mind, the spec isn’t only for the AI—it helps you because the developer preserve oversight and make sure the AI’s work meets the true necessities.

3. Break Duties into Modular Prompts and Context, Not One Huge Immediate

Divide and conquer: Give the AI one centered process at a time moderately than a monolithic immediate with all the things directly.

Skilled AI engineers have discovered that making an attempt to stuff your entire mission (all necessities, all code, all directions) right into a single immediate or agent message is a recipe for confusion. Not solely do you threat hitting token limits; you additionally threat the mannequin dropping focus because of the “curse of directions”—too many directives inflicting it to comply with none of them nicely. The answer is to design your spec and workflow in a modular method, tackling one piece at a time and pulling in solely the context wanted for that piece.

The curse of an excessive amount of context/directions: Analysis has confirmed what many devs anecdotally noticed: as you pile on extra directions or knowledge into the immediate, the mannequin’s efficiency in adhering to every one drops considerably. One examine dubbed this the “curse of directions”, displaying that even GPT-4 and Claude wrestle when requested to fulfill many necessities concurrently. In sensible phrases, should you current 10 bullet factors of detailed guidelines, the AI would possibly obey the primary few and begin overlooking others. The higher technique is iterative focus. Tips from business recommend decomposing complicated necessities into sequential, easy directions as a finest follow. Focus the AI on one subproblem at a time, get that completed, then transfer on. This retains the standard excessive and errors manageable.

Divide the spec into phases or elements: In case your spec doc could be very lengthy or covers numerous floor, take into account splitting it into components (both bodily separate information or clearly separate sections). For instance, you might need a piece for “backend API spec” and one other for “frontend UI spec.” You don’t have to at all times feed the frontend spec to the AI when it’s engaged on the backend, and vice versa. Many devs utilizing multi-agent setups even create separate brokers or subprocesses for every half (e.g., one agent works on database/schema, one other on API logic, one other on frontend—every with the related slice of the spec). Even should you use a single agent, you’ll be able to emulate this by copying solely the related spec part into the immediate for that process. Keep away from context overload: Don’t combine authentication duties with database schema adjustments in a single go, because the DigitalOcean AI information warns. Preserve every immediate tightly scoped to the present aim.

Prolonged TOC/summaries for big specs: One intelligent approach is to have the agent construct an prolonged desk of contents with summaries for the spec. That is primarily a “spec abstract” that condenses every part into just a few key factors or key phrases, and references the place particulars might be discovered. For instance, in case your full spec has a piece on safety necessities spanning 500 phrases, you might need the agent summarize it to: “Safety: Use HTTPS, defend API keys, implement enter validation (see full spec §4.2).” By making a hierarchical abstract within the planning part, you get a hen’s-eye view that may keep within the immediate, whereas the high-quality particulars stay offloaded except wanted. This prolonged TOC acts as an index: The agent can seek the advice of it and say, “Aha, there’s a safety part I ought to have a look at,” and you may then present that part on demand. It’s much like how a human developer skims a top level view after which flips to the related web page of a spec doc when engaged on a particular half.

To implement this, you’ll be able to immediate the agent after writing the spec: “Summarize the spec above into a really concise define with every part’s key factors and a reference tag.” The end result may be a listing of sections with one or two sentence summaries. That abstract might be stored within the system or assistant message to information the agent’s focus with out consuming up too many tokens. This hierarchical summarization strategy is thought to assist LLMs preserve long-term context by specializing in the high-level construction. The agent carries a “psychological map” of the spec.

Make the most of subagents or “abilities” for various spec components: One other superior strategy is utilizing a number of specialised brokers (what Anthropic calls subagents or what you would possibly name “abilities”). Every subagent is configured for a particular space of experience and given the portion of the spec related to that space. As an illustration, you might need a database designer subagent that solely is aware of concerning the knowledge mannequin part of the spec, and an API coder subagent that is aware of the API endpoints spec. The primary agent (or an orchestrator) can route duties to the suitable subagent routinely.

The profit is every agent has a smaller context window to take care of and a extra centered function, which might increase accuracy and permit parallel work on impartial duties. Anthropic’s Claude Code helps this by letting you outline subagents with their very own system prompts and instruments. “Every subagent has a particular objective and experience space, makes use of its personal context window separate from the primary dialog, and has a customized system immediate guiding its conduct,” as their docs describe. When a process comes up that matches a subagent’s area, Claude can delegate that process to it, with the subagent returning outcomes independently.

Parallel brokers for throughput: Operating a number of brokers concurrently is rising as “the subsequent large factor” for developer productiveness. Quite than ready for one agent to complete earlier than beginning one other process, you’ll be able to spin up parallel brokers for non-overlapping work. Willison describes this as “embracing parallel coding brokers” and notes it’s “surprisingly efficient, if mentally exhausting.” The secret is scoping duties so brokers don’t step on one another: One agent codes a function whereas one other writes assessments, or separate elements get constructed concurrently. Orchestration frameworks like LangGraph or OpenAI Swarm might help coordinate these brokers, and shared reminiscence through vector databases (like Chroma) lets them entry widespread context with out redundant prompting.

Single versus multi-agent: When to make use of every

In follow, utilizing subagents or skill-specific prompts would possibly seem like: You preserve a number of spec information (or immediate templates)—e.g., SPEC_backend.md, SPEC_frontend.md—and also you inform the AI, “For backend duties, confer with SPEC_backend; for frontend duties confer with SPEC_frontend.” Or in a instrument like Cursor/Claude, you really spin up a subagent for every. That is actually extra complicated to arrange than a single-agent loop, but it surely mimics what human builders do: We mentally compartmentalize a big spec into related chunks. (You don’t preserve the entire 50-page spec in your head directly; you recall the half you want for the duty at hand, and have a common sense of the general structure.) The problem, as famous, is managing interdependencies: The subagents should nonetheless coordinate. (The frontend must know the API contract from the backend spec, and so on.) A central overview (or an “architect” agent) might help by referencing the subspecs and guaranteeing consistency.

Focus every immediate on one process/part: Even with out fancy multi-agent setups, you’ll be able to manually implement modularity. For instance, after the spec is written, your subsequent transfer may be: “Step 1: Implement the database schema.” You feed the agent the database part of the spec solely, plus any world constraints from the spec (like tech stack). The agent works on that. Then for Step 2, “Now implement the authentication function”, you present the auth part of the spec and perhaps the related components of the schema if wanted. By refreshing the context for every main process, you make sure the mannequin isn’t carrying numerous stale or irrelevant info that might distract it. As one information suggests: “Begin recent: start new classes to clear context when switching between main options.” You’ll be able to at all times remind the agent of crucial world guidelines (from the spec’s constraints part) every time, however don’t shove your entire spec in if it’s not all wanted.

Use in-line directives and code TODOs: One other modularity trick is to make use of your code or spec as an energetic a part of the dialog. As an illustration, scaffold your code with // TODO feedback that describe what must be completed, and have the agent fill them one after the other. Every TODO primarily acts as a mini-spec for a small process. This retains the AI laser centered (“implement this particular perform in response to this spec snippet”), and you may iterate in a decent loop. It’s much like giving the AI a guidelines merchandise to finish moderately than the entire guidelines directly.

The underside line: Small, centered context beats one big immediate. This improves high quality and retains the AI from getting “overwhelmed” by an excessive amount of directly. As one set of finest practices sums up, present “One Process Focus” and “Related information solely” to the mannequin, and keep away from dumping all the things in every single place. By structuring the work into modules—and utilizing methods like spec summaries or subspec brokers—you’ll navigate round context measurement limits and the AI’s short-term reminiscence cap. Bear in mind, a well-fed AI is sort of a well-fed perform: Give it solely the inputs it wants for the job at hand.

4. Construct in Self-Checks, Constraints, and Human Experience

Make your spec not only a to-do record for the agent but in addition a information for high quality management—and don’t be afraid to inject your personal experience.

A superb spec for an AI agent anticipates the place the AI would possibly go flawed and units up guardrails. It additionally takes benefit of what you recognize (area information, edge instances, “gotchas”) so the AI doesn’t function in a vacuum. Consider the spec as each coach and referee for the AI: It ought to encourage the appropriate strategy and name out fouls.

Use three-tier boundaries: GitHub’s evaluation of two,500+ agent information discovered that the simplest specs use a three-tier boundary system moderately than a easy record of don’ts. This offers the agent clearer steering on when to proceed, when to pause, and when to cease:

✅ All the time do: Actions the agent ought to take with out asking. “All the time run assessments earlier than commits.” “All the time comply with the naming conventions within the fashion information.” “All the time log errors to the monitoring service.”

⚠️ Ask first: Actions that require human approval. “Ask earlier than modifying database schemas.” “Ask earlier than including new dependencies.” “Ask earlier than altering CI/CD configuration.” This tier catches high-impact adjustments that may be high-quality however warrant a human examine.

🚫 By no means do: Onerous stops. “By no means commit secrets and techniques or API keys.” “By no means edit node_modules/ or vendor/.” “By no means take away a failing check with out specific approval.” “By no means commit secrets and techniques” was the only commonest useful constraint within the examine.

This three-tier strategy is extra nuanced than a flat record of guidelines. It acknowledges that some actions are at all times secure, some want oversight, and a few are categorically off-limits. The agent can proceed confidently on “All the time” gadgets, flag “Ask first” gadgets for assessment, and hard-stop on “By no means” gadgets.

Encourage self-verification: One highly effective sample is to have the agent confirm its work in opposition to the spec routinely. In case your tooling permits, you’ll be able to combine checks like unit assessments or linting that the AI can run after producing code. However even on the spec/immediate stage, you’ll be able to instruct the AI to double-check (e.g., “After implementing, evaluate the end result with the spec and ensure all necessities are met. Checklist any spec gadgets that aren’t addressed.”). This pushes the LLM to replicate on its output relative to the spec, catching omissions. It’s a type of self-audit constructed into the method.

As an illustration, you would possibly append to a immediate: “(After writing the perform, assessment the above necessities record and guarantee every is glad, marking any lacking ones).” The mannequin will then (ideally) output the code adopted by a brief guidelines indicating if it met every requirement. This reduces the possibility it forgets one thing earlier than you even run assessments. It’s not foolproof, but it surely helps.

LLM-as-a-Choose for subjective checks: For standards which can be onerous to check routinely—code fashion, readability, adherence to architectural patterns—think about using “LLM-as-a-Choose.” This implies having a second agent (or a separate immediate) assessment the primary agent’s output in opposition to your spec’s high quality tips. Anthropic and others have discovered this efficient for subjective analysis. You would possibly immediate “Evaluation this code for adherence to our fashion information. Flag any violations.” The choose agent returns suggestions that both will get included or triggers a revision. This provides a layer of semantic analysis past syntax checks.

Conformance testing: Willison advocates constructing conformance suites—language-independent assessments (usually YAML primarily based) that any implementation should cross. These act as a contract: Should you’re constructing an API, the conformance suite specifies anticipated inputs/outputs, and the agent’s code should fulfill all instances. That is extra rigorous than advert hoc unit assessments as a result of it’s derived instantly from the spec and might be reused throughout implementations. Embrace conformance standards in your spec’s success part (e.g., “Should cross all instances in conformance/api-tests.yaml”).

Leverage testing within the spec: If attainable, incorporate a check plan and even precise assessments in your spec and immediate circulate. In conventional improvement, we use TDD or write check instances to make clear necessities—you are able to do the identical with AI. For instance, within the spec’s success standards, you would possibly say, “These pattern inputs ought to produce these outputs…” or “The next unit assessments ought to cross.” The agent might be prompted to run via these instances in its head or really execute them if it has that functionality. Willison famous that having a strong check suite is like giving the brokers superpowers: They’ll validate and iterate shortly when assessments fail. In an AI coding context, writing a little bit of pseudocode for assessments or anticipated outcomes within the spec can information the agent’s implementation. Moreover, you should utilize a devoted “check agent” in a subagent setup that takes the spec’s standards and constantly verifies the “code agent’s” output.

Convey your area information: Your spec ought to replicate insights that solely an skilled developer or somebody with context would know. For instance, should you’re constructing an ecommerce agent and you recognize that “merchandise” and “classes” have a many-to-many relationship, state that clearly. (Don’t assume the AI will infer it—it may not.) If a sure library is notoriously difficult, point out pitfalls to keep away from. Basically, pour your mentorship into the spec. The spec can include recommendation like “If utilizing library X, be careful for reminiscence leak problem in model Y (apply workaround Z).” This stage of element is what turns a median AI output into a really strong resolution, since you’ve steered the AI away from widespread traps.

Additionally, in case you have preferences or fashion tips (say, “use purposeful elements over class elements in React”), encode that within the spec. The AI will then emulate your fashion. Many engineers even embrace small examples within the spec (as an example, “All API responses needs to be JSON, e.g., {“error”: “message”} for errors.”). By giving a fast instance, you anchor the AI to the precise format you need.

Minimalism for easy duties: Whereas we advocate thorough specs, a part of experience is realizing when to maintain it easy. For comparatively easy, remoted duties, an overbearing spec can really confuse greater than assist. Should you’re asking the agent to do one thing easy (like “middle a div on the web page”), you would possibly simply say, “Make sure that to maintain the answer concise and don’t add extraneous markup or kinds.” No want for a full PRD there. Conversely, for complicated duties (like “implement an OAuth circulate with token refresh and error dealing with”), that’s if you get away the detailed spec. A superb rule of thumb: Alter spec element to process complexity. Don’t underspec a tough downside (the agent will flail or go off-track), however don’t overspec a trivial one (the agent would possibly get tangled or dissipate context on pointless directions).

Preserve the AI’s “persona” if wanted: Typically, a part of your spec is defining how the agent ought to behave or reply, particularly if the agent interacts with customers. For instance, if constructing a buyer help agent, your spec would possibly embrace tips like “Use a pleasant {and professional} tone” and “Should you don’t know the reply, ask for clarification or supply to comply with up moderately than guessing.” These sorts of guidelines (usually included in system prompts) assist preserve the AI’s outputs aligned with expectations. They’re primarily spec gadgets for AI conduct. Preserve them constant and remind the mannequin of them if wanted in lengthy classes. (LLMs can “drift” in fashion over time if not stored on a leash.)

You stay the exec within the loop: The spec empowers the agent, however you stay the last word high quality filter. If the agent produces one thing that technically meets the spec however doesn’t really feel proper, belief your judgement. Both refine the spec or instantly modify the output. The beauty of AI brokers is that they don’t get offended—in the event that they ship a design that’s off, you’ll be able to say, “Truly, that’s not what I meant, let’s make clear the spec and redo it.” The spec is a residing artifact in collaboration with the AI, not a one-time contract you’ll be able to’t change.

Simon Willison humorously likened working with AI brokers to “a really bizarre type of administration” and even “getting good outcomes out of a coding agent feels uncomfortably near managing a human intern.” You have to present clear directions (the spec), guarantee they’ve the mandatory context (the spec and related knowledge), and provides actionable suggestions. The spec units the stage, however monitoring and suggestions throughout execution are key. If an AI was a “bizarre digital intern who will completely cheat should you give them an opportunity,” the spec and constraints you write are the way you stop that dishonest and preserve them on process.

Right here’s the payoff: A superb spec doesn’t simply inform the AI what to construct; it additionally helps it self-correct and keep inside secure boundaries. By baking in verification steps, constraints, and your hard-earned information, you drastically improve the percentages that the agent’s output is appropriate on the primary attempt (or a minimum of a lot nearer to appropriate). This reduces iterations and people “Why on Earth did it do this?” moments.

5. Check, Iterate, and Evolve the Spec (and Use the Proper Instruments)

Consider spec writing and agent constructing as an iterative loop: check early, collect suggestions, refine the spec, and leverage instruments to automate checks.

The preliminary spec shouldn’t be the top—it’s the start of a cycle. The perfect outcomes come if you frequently confirm the agent’s work in opposition to the spec and modify accordingly. Additionally, trendy AI devs use varied instruments to help this course of (from CI pipelines to context administration utilities).

Steady testing: Don’t wait till the top to see if the agent met the spec. After every main milestone and even every perform, run assessments or a minimum of do fast guide checks. If one thing fails, replace the spec or immediate earlier than continuing. For instance, if the spec mentioned, “Passwords have to be hashed with bcrypt” and also you see the agent’s code storing plain textual content, cease and proper it (and remind the spec or immediate concerning the rule). Automated assessments shine right here: Should you offered assessments (or write them as you go), let the agent run them. In lots of coding agent setups, you’ll be able to have an agent run npm check or related after ending a process. The outcomes (failures) can then feed again into the subsequent immediate, successfully telling the agent “Your output didn’t meet spec on X, Y, Z—repair it.” This type of agentic loop (code > check > repair > repeat) is extraordinarily highly effective and is how instruments like Claude Code or Copilot Labs are evolving to deal with bigger duties. All the time outline what “completed” means (through assessments or standards) and examine for it.

Iterate on the spec itself: Should you uncover that the spec was incomplete or unclear (perhaps the agent misunderstood one thing otherwise you realized you missed a requirement), replace the spec doc. Then explicitly resync the agent with the brand new spec: “I’ve up to date the spec as follows… Given the up to date spec, modify the plan or refactor the code accordingly.” This manner the spec stays the only supply of reality. It’s much like how we deal with altering necessities in regular dev, however on this case you’re additionally the product supervisor to your AI agent. Preserve model historical past if attainable (even simply through commit messages or notes), so you recognize what modified and why.

Make the most of context administration and reminiscence instruments: There’s a rising ecosystem of instruments to assist handle AI agent context and information. As an illustration, retrieval-augmented technology (RAG) is a sample the place the agent can pull in related chunks of information from a information base (like a vector database) on the fly. In case your spec is large, you can embed sections of it and let the agent retrieve probably the most related components when wanted, as an alternative of at all times offering the entire thing. There are additionally frameworks implementing the Mannequin Context Protocol (MCP), which automates feeding the appropriate context to the mannequin primarily based on the present process. One instance is Context7 (context7.com), which might auto-fetch related context snippets from docs primarily based on what you’re engaged on. In follow, this would possibly imply the agent notices you’re engaged on “fee processing” and it pulls the funds part of your spec or documentation into the immediate. Take into account leveraging such instruments or organising a rudimentary model (even a easy search in your spec doc).

Parallelize fastidiously: Some builders run a number of agent cases in parallel on completely different duties (as talked about earlier with subagents). This may velocity up improvement (e.g., one agent generates code whereas one other concurrently writes assessments, or two options are constructed concurrently). Should you go this route, make sure the duties are really impartial or clearly separated to keep away from conflicts. (The spec ought to notice any dependencies.) For instance, don’t have two brokers writing to the identical file directly. One workflow is to have an agent generate code and one other assessment it in parallel, or to have separate elements constructed that combine later. That is superior utilization and might be mentally taxing to handle. (As Willison admitted, operating a number of brokers is surprisingly efficient, if mentally exhausting!) Begin with at most 2–3 brokers to maintain issues manageable.

Model management and spec locks: Use Git or your model management of selection to trace what the agent does. Good model management habits matter much more with AI help. Commit the spec file itself to the repo. This not solely preserves historical past, however the agent may even use git diff or blame to know adjustments. (LLMs are fairly able to studying diffs.) Some superior agent setups let the agent question the VCS historical past to see when one thing was launched—surprisingly, fashions might be “fiercely competent at Git.” By holding your spec within the repo, you permit each you and the AI to trace evolution. There are instruments (like GitHub Spec Equipment talked about earlier) that combine spec-driven improvement into the Git workflow—as an example, gating merges on up to date specs or producing checklists from spec gadgets. Whilst you don’t want these instruments to succeed, the takeaway is to deal with the spec like code: Preserve it diligently.

Price and velocity concerns: Working with massive fashions and lengthy contexts might be gradual and costly. A sensible tip is to make use of mannequin choice and batching well. Maybe use a less expensive/quicker mannequin for preliminary drafts or repetitions, and reserve probably the most succesful (and costly) mannequin for remaining outputs or complicated reasoning. Some builders use GPT-4 or Claude for planning and demanding steps, however offload less complicated expansions or refactors to a neighborhood mannequin or a smaller API mannequin. If utilizing a number of brokers, perhaps not all should be high tier; a test-running agent or a linter agent may very well be a smaller mannequin. Additionally take into account throttling context measurement: Don’t feed 20K tokens if 5K will do. As we mentioned, extra tokens can imply diminishing returns.

Monitor and log all the things: In complicated agent workflows, logging the agent’s actions and outputs is important. Verify the logs to see if the agent is deviating or encountering errors. Many frameworks present hint logs or permit printing the agent’s chain of thought (particularly should you immediate it to suppose step-by-step). Reviewing these logs can spotlight the place the spec or directions might need been misinterpreted. It’s not not like debugging a program—besides the “program” is the dialog/immediate chain. If one thing bizarre occurs, return to the spec/directions to see if there was ambiguity.

Study and enhance: Lastly, deal with every mission as a studying alternative to refine your spec-writing talent. Possibly you’ll uncover {that a} sure phrasing constantly confuses the AI, or that organizing spec sections in a sure method yields higher adherence. Incorporate these classes into the subsequent spec. The sphere of AI brokers is quickly evolving, so new finest practices (and instruments) emerge always. Keep up to date through blogs (like those by Simon Willison, Andrej Karpathy, and so on.), and don’t hesitate to experiment.

A spec for an AI agent isn’t “write as soon as, completed.” It’s a part of a steady cycle of instructing, verifying, and refining. The payoff for this diligence is substantial: By catching points early and holding the agent aligned, you keep away from expensive rewrites or failures later. As one AI engineer quipped, utilizing these practices can really feel like having “a military of interns” working for you, however it’s important to handle them nicely. A superb spec, constantly maintained, is your administration instrument.

Keep away from Widespread Pitfalls

Earlier than wrapping up, it’s value calling out antipatterns that may derail even well-intentioned spec-driven workflows. The GitHub examine of two,500+ agent information revealed a stark divide: “Most agent information fail as a result of they’re too imprecise.” Listed below are the errors to keep away from:

Imprecise prompts: “Construct me one thing cool” or “Make it work higher” provides the agent nothing to anchor on. As Baptiste Studer places it: “Imprecise prompts imply flawed outcomes.” Be particular about inputs, outputs, and constraints. “You’re a useful coding assistant” doesn’t work. “You’re a check engineer who writes assessments for React elements, follows these examples, and by no means modifies supply code” does.

Overlong contexts with out summarization: Dumping 50 pages of documentation right into a immediate and hoping the mannequin figures it out hardly ever works. Use hierarchical summaries (as mentioned in precept 3) or RAG to floor solely what’s related. Context size shouldn’t be an alternative to context high quality.

Skipping human assessment: Willison has a private rule—“I gained’t commit code I couldn’t clarify to another person.” Simply because the agent produced one thing that passes assessments doesn’t imply it’s appropriate, safe, or maintainable. All the time assessment crucial code paths. The “home of playing cards” metaphor applies: AI-generated code can look strong however collapse underneath edge instances you didn’t check.

Conflating vibe coding with manufacturing engineering: Fast prototyping with AI (“vibe coding”) is nice for exploration and throwaway initiatives. However delivery that code to manufacturing with out rigorous specs, assessments, and assessment is asking for hassle. I distinguish “vibe coding” from “AI-assisted engineering”—the latter requires the self-discipline this information describes. Know which mode you’re in.

Ignoring the “deadly trifecta”: Willison warns of three properties that make AI brokers harmful: velocity (they work quicker than you’ll be able to assessment), nondeterminism (similar enter, completely different outputs), and value (encouraging nook slicing on verification). Your spec and assessment course of should account for all three. Don’t let velocity outpace your potential to confirm.

Lacking the six core areas: In case your spec doesn’t cowl instructions, testing, mission construction, code fashion, git workflow, and limits, you’re seemingly lacking one thing the agent wants. Use the six-area guidelines from part 2 as a sanity examine earlier than handing off to the agent.

Conclusion

Writing an efficient spec for AI coding brokers requires strong software program engineering rules mixed with adaptation to LLM quirks. Begin with readability of objective and let the AI assist increase the plan. Construction the spec like a severe design doc, masking the six core areas and integrating it into your toolchain so it turns into an executable artifact, not simply prose. Preserve the agent’s focus tight by feeding it one piece of the puzzle at a time (and take into account intelligent ways like abstract TOCs, subagents, or parallel orchestration to deal with large specs). Anticipate pitfalls by together with three-tier boundaries (at all times/ask first/by no means), self-checks, and conformance assessments—primarily, educate the AI how you can not fail. And deal with the entire course of as iterative: use assessments and suggestions to refine each the spec and the code constantly.

Observe these tips and your AI agent will probably be far much less more likely to “break down” underneath massive contexts or wander away into nonsense.

Comfortable spec-writing!