Fortin frames Copilot skill as a ladder rather than a binary "use it or not" state. Level one (Explorer) is pre-built prompts and web-search-style questions in apps like Outlook, which yield generic results because the user provides no context. Levels escalate through structured prompting, app integration with file and brand awareness, custom agents, and finally Co-Work, which chains multiple steps across apps autonomously. The framing positions Copilot mastery as a skill progression, not a feature tour [Source 1].

The Prompter level introduces Goal, Context, Expectations, and Source. In the demo, a weekly sales report PDF from a fictional pizzeria is summarized two ways: a basic "summarize this report" prompt versus a GCES-structured prompt that states targets (47k weekly, food costs under 30%), desired outputs (three things that went right, three that went wrong, what needs attention first), and the source. The structured result includes an actionable table, assigns owners, and cites specific PDF sections (section four, the oven), demonstrating that prompt structure directly controls output usefulness [Source 1].

Level three covers Copilot in PowerPoint, Excel, and Word. Key demonstrated techniques: enable "allow editing" mode so Copilot edits the file directly, attach files or OneDrive content, select a brand so the deck matches company assets, and explicitly pick a model (Claude Opus 5 or GPT 5.6 in his environment) instead of auto. Fortin generated an eight-slide presentation from a PDF in about nine minutes, then issued a follow-up prompt that pulled the latest email from Outlook into a new slide matching the deck style, showing cross-app data access [Source 1].

Level four distinguishes pre-built Microsoft agents from custom agents. The Analyst agent ran a 17-step comparison of two weekly sales reports and produced week-over-week analysis, charts, and a comparison table. Fortin argues agents outperform ordinary prompts for their specialty because they are trained and instructed for that single task. He also shows custom agents: a "PowerPoint Guru" that prepares presentation question banks, deck summaries, and top risks, and a LinkedIn post agent that generates three hook ideas and a post from a pasted transcript. Custom agents are configured with plain-language instructions in the agent builder [Source 1].

The capstone demonstration uses Copilot Co-Work to migrate roughly 70 course testimonials from templated Thinkific notification emails into a structured Excel spreadsheet, which Fortin then imported into the Senja testimonial platform. He provided an example image of the email format so Co-Work understood the extraction pattern, and ran the workflow since January 1, 2026. The run cost approximately 783 credits (about $7.83). Fortin estimates the equivalent manual copy-paste would have taken 3 to 4 hours, framing this as a small-cost, high-return automation pattern [Source 1].

Several features require a paid M365 Copilot license: level 3.5 follow-up slide generation, agents, and Co-Work. Fortin notes admins can turn off agents in tenant policy, and that Copilot buttons appear across 365 apps depending on the license label. Model availability also varies by tenant, with Claude and GPT offered alongside Microsoft models in his setup. These constraints matter for organizations planning Copilot rollout, since feature availability is not uniform [Source 1].

• Single-source tutorial bias: The brief derives from one YouTube video by an educator who sells a Copilot course, so feature descriptions may be optimized for engagement and conversion rather than critical evaluation.
• Feature availability uncertainty: Agents, Co-Work, model selection, and brand integration vary by license tier, tenant rollout phase, and organizational policy; a free or standard license will not match the demo.
• Transcript quality: Auto-generated captions may contain terminology or model-name errors; model names like GPT 5.6 and Claude Opus 5 reflect the speaker's environment and date.
• Self-reported time savings: The "3 hours saved in 5 minutes" claim is the speaker's estimate of a single workflow and generalizes only to similar high-volume, template-driven extraction tasks.
• Cost basis: The 783-credit figure is specific to his mail volume and tenant; credit pricing and Co-Work availability may have changed since recording.
• Not covered: The video does not cover governance, data security defaults, admin configuration for Co-Work, or enterprise rollout considerations.

• 1

  Audit current Copilot usage against the five levels — Identify which level describes most of Steve's Copilot interactions and target the next level up as a deliberate practice goal.

• 2

  Adopt the GCES prompting framework — Rewrite the next three Copilot prompts used in real work with explicit Goal, Context, Expectations, and Source sections, and compare output quality against an unstructured prompt.

• 3

  Test model selection in Copilot apps — Instead of leaving the model on auto, experiment with explicit Claude or GPT selection for a presentation or analysis task and note quality and speed differences.

• 4

  Build one custom agent — Pick a recurring task, such as presentation prep or transcript-to-post drafting, configure an agent with clear instructions, and run it for two weeks.

• 5

  Evaluate a Co-Work automation candidate — Identify one high-volume, template-driven data task in Steve's workflow and estimate the credit cost and manual time it would replace before running it.

• 6

  Review Fortin's prompt templates and ebook — The video references a downloadable ebook with every prompt used, a reusable resource for copying GCES-style prompts.

• [1] Every Level of Copilot Explained in 18 Minutes — YouTube (Tier 2)

Synthesized by agent Knight (residential PC) from a single YouTube source. Confidence Medium-High.