8 hour customer workshop
Reduce token spend without losing quality. Set enterprise spending limits.
Input, output, cached tokens, agent loops, MCP schemas, model routing.
Prompt compression, output control, context hygiene, Ask/Edit/Agent routing.
Enterprise budgets, model policy, instruction scope, monthly review loop.
Use docs.github.com/copilot for supported feature behavior, billing controls, and admin policy.
| Time | Block | Output |
|---|---|---|
| 00:00-00:30 | Why tokens matter + fast wins | Shared cost model |
| 00:30-02:00 | Prompt, language, output labs | Terse prompt templates |
| 02:00-03:15 | Context + always-on files | Instruction pruning diff |
| 03:15-04:15 | Workflow + modes | Mode routing cheat sheet |
| 04:15-05:15 | MCP/tool costs + data | MCP audit list |
| 05:15-06:15 | Practical setup | Repo setup checklist |
| 06:15-07:10 | Model/pricing + governance | Customer admin rollout |
| 07:10-07:40 | Outcome per token | Plan/execute/verify loop |
| 07:40-08:00 | Capstone | 30 day token plan |
Move horizontally between chapters. Move vertically through each chapter's slides and labs.
Press Esc for overview. Press S for speaker notes.
Quick Start
Cut unwanted output, reduce repeated context, then choose models and tools by task.
applyTo./chronicle cost tips & /chronicle improve in Copilot CLI.Set a concise response default.
Remove context repeated on every turn.
Choose Ask, Edit, or Agent for the task.
Budgets and model policy cap spend.
Plan, route, verify, close.
Identify your top 3 token leaks before learning techniques.
Do: mark each as green/yellow/red: output verbosity, loaded instructions, open tabs, Agent use, loaded tools, model pinning, long chat history.
Deliverable: personal baseline + one "fix today" action.
Time: 10 minutes.
Part 1
Tokens drive cost, speed, limits, and context capacity.
Sure, I'd be happy to help! uses around 10 tokens without answering the question.
Output tokens usually cost materially more than input tokens in vendor pricing examples.
Context window
├─ System prompt
├─ copilot-instructions.md / AGENTS.md
├─ File context and open tabs
├─ Conversation history
├─ MCP tool schemas
└─ Your typed prompt
A 20-word prompt can be sent with thousands of context tokens.
One issue-to-PR run is not one request.
Plan, search, read files, edit, run tests, inspect failures, retry.
Later steps carry retained context and tool results. Caching and compaction affect the cost.
Separate visible prompt from hidden context.
Do: choose one recent Copilot task. List likely context sources: open files, referenced files, instructions, history, tools, output.
Discuss: which source was most controllable?
Time: 15 minutes.
Part 2.1
Make the request clear before making it shorter. Keep technical terms and constraints exact.
Ordinary concise language is the default. Shorthand is optional.
Redundant wording and pleasantries that add no task information.
Technical terms, code, filenames, negation, conditions, and acceptance checks.
[thing] [action] [reason]. [next step].
Keep full sentences when fragments hide who does what.
| Wordy | Concise |
|---|---|
| Could you please review this pull request and let me know if there are any issues? | Review PR. Flag issues. |
| Can you explain what this error message means and how I should fix it? | Explain this error and how to fix it. |
| Level | Style | Use when |
|---|---|---|
| Concise prose | Plain, complete wording | Default |
| Fragments | Shared terms, short clauses | Actors and conditions stay clear |
| Dense shorthand | Abbreviations, arrows | Only if readers can decode it reliably |
POST /api/users
Validate:
- name: string, required
- email: string, required, valid
400 on validation fail
201 on success, return created user
Save to DB
Bullets and key-value pairs make distinct requirements easier to check.
Keep positive numbers, double them, and return their sum.
fn(nums) -> filter(>0) -> map(*2) -> sum
Like getUserById but for emails. 404 if missing.
| Verbose procedure | Compressed invariant |
|---|---|
| Make sure every SQL query parameterizes values... | SQL: parameterized queries only. No concatenation. |
| Please write tests for any new code... | New code -> tests. Cover happy + error. |
Compare concise prose, fragments, and shorthand for one real prompt.
Do: preserve file names, line numbers, constraints, done condition. Drop only fluff.
Compare: which version makes the task easiest to understand without losing requirements?
Time: 20 minutes.
A: Could you take a look at auth and improve error handling?
B: File: src/auth/login.ts.
Bug: null user causes .email crash.
Fix: null guard before .email.
Test: add null-user case.
Done: targeted test passes.
Run: use separate sessions for A and B. Compare file reads, clarifications, and results. B adds requirements, so this tests specificity, not compression alone.
Part 2.2
Compare token counts, not character counts.
| Language | Sentence | Tokens | vs English |
|---|---|---|---|
| English | I met a huge dog | 5 | 1.0x |
| Spanish | Conoci a un perro enorme | 8 | 1.6x |
| Chinese | Chinese equivalent | 11 | 2.2x |
| Japanese | Japanese equivalent | 11 | 2.2x |
| Russian | Russian equivalent | 14 | 2.8x |
English uses fewer tokens in these examples. Test equivalent prompts with the target tokenizer before generalizing.
Disprove or confirm language assumptions with actual token counts.
Do: pick one prompt. Translate to your strongest non-English language. Compare token counts in tokenizer tool.
Deliverable: measured token counts and any clarity tradeoffs.
Time: 20 minutes.
Part 2.3
Control which files, instructions, and history get sent.
.github/copilot-instructions.md, AGENTS.md, CLAUDE.md, and similar files can become persistent context.
Duplicate rules add context twice when both files are loaded.
Delete duplicate rules, stale notes, and facts visible in code. Keep project-specific pitfalls.
The reusable response template lives in Practical Setup. No token-count estimate is needed to decide whether a rule is relevant.
---
applyTo: "src/api/**/*.ts"
---
API conventions:
- Routes in src/api/routes/. Handlers thin, logic in services/.
- Validate with zod.
- Errors via Result<T,E>, never throw.
Load path-specific guidance only for matching files.
Configure Content Exclusion for sensitive files, generated bundles, large data files, and regulated paths.
Important: this is a privacy/policy control, not a general token setting. It does not apply to Copilot CLI, cloud agent, or IDE Agent mode. Check current official support.
Find always-on context that should become scoped or on-demand.
Do: inspect repo instructions, AGENTS/CLAUDE files, open tabs, generated files, MCP config.
Deliverable: three-column list: keep always-on, move to applyTo, move on-demand/delete.
Time: 30 minutes.
Part 2.4
Ask for the answer format you need. Skip unwanted explanation.
Code only, no explanation.
Use when the team understands the change. Keep errors, missing requirements, and failed checks visible.
| Instruction | Use when |
|---|---|
| Answer in one sentence | Quick decision or explanation |
| 3 bullets max | Scan-friendly summary |
| Reply as JSON | Machine-readable extraction |
| Yes/no, then one line why | Review gate |
| Diff only | Patch inspection |
Lead with the answer. Keep explanations proportional to the task, with verification evidence and blockers intact.
Use the single starter template, then add relevant project rules.
A: Add input validation to processOrder().
B: Add input validation to processOrder().
Code only. No explanation. Minimal diff.
Measure: acceptance checks, full token usage, billed cost, time, and retries, not only response length.
Time: 15 minutes.
Practice asking for bounded output.
Prompts: "3 bullets max", "table only", "JSON only", "one-line verdict + risk".
Deliverable: team snippet library for common Copilot outputs.
Part 2.5
Choose a lower-overhead workflow before shortening individual prompts.
feat: add password reset via settings page
Body only when "why" is not obvious.
L42: bug: user can be null. Add guard before .email.
One-line, actionable, severity-coded.
Auto is a good baseline: supported Auto pool, org policy-aware, discount on eligible paid-plan Chat usage per docs.
Auto does not mean "escalate into every premium model." Pin premium deliberately.
Target model: GPT-5.5.
Guide: <official prompting guide URL>
Files: .github/copilot-instructions.md, .github/instructions/*.md
Adapt prompts to guide. Preserve behavior. Reduce rework.
Show diff only.
/chronicleCopilot CLI session history (not VS Code). Use /chronicle cost tips for token spend and /chronicle improve for recurring confusion.
VS Code local extension. Finds anti-patterns, context health, skill opportunities.
Stop using Agent Mode for one-shot work.
Do: classify 20 sample tasks as Ask/Edit/Agent/Coding Agent. Defend choice.
Deliverable: team mode-routing cheat sheet.
Time: 20 minutes.
Prevent expensive-model pinning by default.
Do: map tasks to Auto, included/lower-cost, standard, premium. Include reasoning-effort choice if available.
Deliverable: model escalation rules.
Part 2.6
More context can make agents worse and cost more.
| Finding | Observed impact |
|---|---|
| LLM-generated context files | Hurt in 5/8 settings |
| Average correctness | Down about 2% |
| Token cost | Up 20-23% |
| Reasoning overhead | Up about 22% in cited setup |
Repeats facts the agent can find in code.
Buries important rules in long context.
Steers the agent toward outdated tools.
Makes project-specific pitfalls harder to spot.
| Keep | Delete |
|---|---|
Use uv instead of pip | This is a Python project |
| DB migrations must run in order | We use PostgreSQL |
| Do not refactor auth module; audit pending | We use JWT auth |
| Deploy requires VPN | Main branch protected |
Start almost empty.
Agent trips on something -> add one line.
Root cause fixed -> delete that line.
Remove instructions when the underlying problem is fixed.
Remove unnecessary context; check that needed guidance remains.
Do: mark each instruction as discoverable, style, landmine, stale, duplicate. Delete or scope everything except true always-on landmines.
Deliverable: a pruned instruction draft with necessary rules intact.
Time: 35 minutes.
Part 2.7
Loaded tool definitions add input, even when the tool is unused.
/context
System/Tools: MCPs + instructions + system prompt
Messages: conversation history
Free Space: remaining context
Inspect loaded tools. Installed server counts alone do not capture deferred definitions or cache behavior.
| Component | Approx cost |
|---|---|
| Name + description | 20-50 tokens |
| Simple parameter schema | 30-80 tokens |
| Complex parameter schema | 100-300 tokens |
| Total per tool | 100-500 tokens |
10 MCP servers
x 5 tools each
x 200 tokens/tool
= 10,000 tokens per step
15 agent steps = 150,000 schema tokens
Hypothetical full-catalog loading, not measured billing. Deferred definitions and caching change the result.
Rust CLI proxy filters noisy shell output: tests, git diff, grep, logs, file listings.
rtk init --copilot
YAML-extensible command filters with local savings stats and team-maintained rules.
snip init --agent copilot
Pick one output filter layer per command path. Do not stack RTK and snip by default.
| Layer | Tool | Reduces |
|---|---|---|
| Workflow turns | CodeAct | Repeated model-tool loops |
| Command output | RTK or snip | Verbose shell results |
| Command choice | minimal-context-tools | Broad search/read behavior |
| Codebase orientation | Graphify | Repeated file rereads |
| Visibility | Chronicle / Coach | Waste you would miss |
One large plugin can account for most of the System/Tools context.
--namespace appservice --namespace cosmos --namespace keyvault --namespace storage
Load the services each role needs. Avoid "all tools" mode unless the task requires it.
Remove hidden schema overhead.
Do: inspect loaded definitions and metadata, enabled servers, global vs workspace scope, owner, and frequency of use.
Deliverable: keep/disable/scope table + restart plan.
Time: 30 minutes.
Define approved MCP servers by persona: developer, data, platform, security, support.
Deliverable: default MCP profile per team + exception process for large plugins.
Part 3
Compare measured effects with their baseline, workload, and limits.
| Study | Control | Result |
|---|---|---|
| Upstream | Answer concisely. | 3% median output reduction; 10 prompts, one run |
| JetBrains | Stock Claude Code | 8.5% fewer output tokens; 82 paired tasks |
| JetBrains total cost | Same full run | 11.6% higher; long-context outlier |
Different controls, not universal savings. Sources, versions, and quality limits.
Answer concisely.Repeat paired tasks. Track acceptance checks, token buckets, cost, time, and retries. This guide has not run its own paid benchmark.
Pick changes by impact, effort, risk.
Do: rate each technique 1-5 for impact, effort, adoption friction, governance need.
Deliverable: team top 5 and "do not adopt" list.
Time: 25 minutes.
Part 4
Turn techniques into repo and team defaults.
applyTo.copilot-setup-steps.yml for Coding Agent.Lead with the answer. Keep explanations proportional to the task.
Skip greetings, repeated summaries, and unsolicited follow-up offers.
Preserve technical terms, negation, conditions, numbers, units, code, and commands.
Use full sentences when brevity would make an action ambiguous.
For code changes, report what changed and the relevant test result or blocker.
Canonical template: Practical Setup.
# .github/copilot-setup-steps.yml
steps:
- name: Install dependencies
run: npm ci
- name: Build
run: npm run build
Illustrative steps, not a complete workflow file. Deterministic setup can avoid repeated dependency discovery.
Bug: login fails when email contains "+".
File: src/auth/login.ts, validateEmail() L42.
Fix: URL-encode email before OAuth provider call.
Test: add user+tag@example.com case.
Done: targeted test passes.
| Week | Habit |
|---|---|
| 1 | Record a baseline; prune irrelevant instructions |
| 2 | Use concise requests and task-specific response formats |
| 3 | Audit loaded tools, command output, and context |
| 4 | Compare accepted outcomes, total cost, time, and retries |
{
"chat.agent.maxRequests": 10,
"github.copilot.chat.agent.model": "auto"
}
Cap runaway sessions carefully. Increase only when task requires it.
model
mode
agent/profile
active MCP/tools
output filter
repo instructions
Choose settings before the session starts. Keeping them stable helps preserve cached context.
Create a repo-ready token optimization PR draft.
Do: draft the changes your repo needs: copilot-instructions.md, scoped guidance, Coding Agent setup, or a precise issue snippet. Skip artifacts already adequate.
Deliverable: branch or patch proposal.
Time: 45 minutes.
Part 4.2
Separate Copilot docs, UBB framing, and vendor token pricing.
| View | Answers |
|---|---|
| GitHub Copilot docs | Plan availability, model access, Auto behavior, published signals |
| UBB framing | AI-credit budgets and governance after cutover |
| Vendor API pricing | Input vs output intuition, not Copilot billing table |
| Effort | Use for |
|---|---|
| Low | High-volume simple chat/classification |
| Medium | Typical coding and tool-heavy work where supported |
| High/max | Architecture, security, novel decomposition |
Use only on models that expose this control.
Define when users may pin premium models and when Auto is required.
Do: map workflows to default model lane, escalation trigger, expected value, rollback signal.
Deliverable: one-page model policy draft.
Time: 30 minutes.
Part 4.3
Admin controls cap spend. Prompt habits improve efficiency.
Prompt compression does not cap spend. It reduces waste inside allowed usage.
Set enterprise/org/cost-center budgets. Enable alerts early. Enable stop-usage after reporting is trusted. Review monthly.
$0 budget: no usage-based features.| Question | Decision |
|---|---|
| Which workflows need premium? | Enable narrowly |
| Which teams create measurable value? | Assign higher budgets |
| Which users can stay on Auto? | Keep default path cheap |
| What rollback signal? | Usage up, value flat |
Useful for different model policies or billing boundaries when org structure already matches groups.
Costs: admin overhead, license complexity, user sprawl, SCIM/cost-center constraints.
Design budget rollout for one enterprise account.
Do: choose pilot org, cost center, baseline budget, alert threshold, stop-usage rule, user exceptions.
Deliverable: budget operating model.
Time: 35 minutes.
Agent sessions spike AI-credit usage 4x in one team. Delivery output unchanged.
Decide: budget action, model policy action, instruction/context action, MCP audit action, comms plan.
Time: 25 minutes.
Part 4.4
Optimize for accepted work, not the shortest prompt.
A longer plan can cost less than a short prompt that leads to the wrong implementation.
| Finding | Implication |
|---|---|
| Agentic coding can consume far more than chat | Do not extrapolate simple-chat cost |
| Same task can vary widely across runs | Use budgets and stop rules |
| More tokens do not guarantee accuracy | Optimize loop quality |
| Input dominates agentic cost | Context hygiene matters |
| Skill | Potential benefit |
|---|---|
| Brainstorming | Prevents early lock-in |
| Planning | Reduces guessing during execution |
| Greenness | Avoids debugging unknown baseline failures |
| Verification | Prevents false completion |
| Impeccable close | Prevents review churn |
| Branch-close discipline | Stops stale context carryover |
Superpowers, planning, TDD, branch finish.
agent-toolkit patterns for requirements, plans, and removing unnecessary work.
Secrets, cloud, DB, Docker, dependency, and retry-stop cautions.
Browser evidence, writing review, QA depth on demand.
Do not install every skill. Load only what changes the next action.
Stars show attention, not quality. Judge by whether the skill changes the next agent action.
Capstone
Choose the changes, owners, and dates for a customer rollout.
| Week | Team action | Admin action |
|---|---|---|
| 1 | Output defaults + Ask Mode | Usage baseline |
| 2 | Instruction pruning + scoped rules + Markdown conversion | Budget pilot |
| 3 | MCP audit + output filter + Graphify pilot | Model access review |
| 4 | Plan-first execution + verification/close discipline | Monthly outcome-per-token review |
Create an action plan for one customer.
Sections: repo changes, user habits, MCP/tool policy, model policy, budgets, reporting, owner, due date.
Deliverable: 30 day plan ready for customer sponsor.
Time: 30 minutes.
applyTo files