Aicoursescanada

Every build on this page is illustrative. These are composite scenarios describing the shape of our work, written so you can see our method. They are not client accounts, and no numbers here should be read as verified results.

Build 1

A municipal service desk

Sector — municipal · Format — evening cohort + desk rules · Length — 5 weeks

The situation

A front counter and an email queue sat on the same small team. Scripts existed. What did not exist was a shared way to draft a reply when the script ran out, without dropping a name, an address, or a file number into a public assistant. New staff copied old threads. Older staff rewrote every letter from a blank page.

Managers wanted speed and a consistent voice. Records staff wanted a trail. The teaching problem was not “how to prompt.” It was how a shift could use a model for wording while keeping identity in the case system, where it already belonged.

What we built

  • A classification strip on the desk. Four buckets, in the language the team already used, with examples from their queue (rewritten so no person could be recognized).
  • Answer frames, not canned letters. Openings, middle asks, and closings that could be completed from the case file by a human.
  • A red-line card. What never leaves the case system: identifiers, health-adjacent notes, anything still under investigation.
  • A shift briefing. Ten minutes at handover covering new templates and one failure example from the week.
  • A light log. Tool, task class, initials — enough to answer “did we put that in a chat?” without building a surveillance habit.

The curriculum arc

Week 1, baseline and classification. Week 2, frames for the three most common request types. Week 3, the red-line drills and de-identification. Week 4, shift briefing and the log. Week 5, review of live (still de-identified) drafts, line by line.

What changed in the work

Drafting time stopped being a private craft. People could point at a frame and a rule instead of a personality. The voice of the desk got duller in a useful way: fewer flourishes, fewer extra facts in the closing paragraph. Managers could see which request types still needed a specialist, because those were the ones the frames refused to cover.

What stayed hard

Angry threads and anything that mixed several files still needed a senior officer. A model is fluent under pressure in the worst sense: it will soothe in a sentence that accidentally confirms something. We left that class of mail off the tool, on purpose, and said so in the binder.

Build 2

A 30-person trades contractor

Sector — field operations · Format — morning stream + studio hours · Length — 6 weeks

The situation

Estimating lived in a few people’s heads and in a folder of old PDFs. Work orders were typed from voice notes recorded in a parking lot. Supplier mail was polite and slow. The owner wanted the office to draft faster without a model inventing a linear-foot count or a client’s site address appearing in a consumer tool.

The people who knew the assemblies were often on site. The people at the desk needed language and structure, not a lecture on neural nets. We treated the model as a clerk for wording, never as a quantity surveyor.

What we built

  • A quote frame with locked fields. Scope and exclusions in prose; quantities and prices filled only from the estimate sheet.
  • A small knowledge base of standard assemblies. Written by the leads, dated, with a “do not invent a substitute” instruction.
  • Vendor correspondence sets. Lead times, shortages, and change orders in a voice that still sounded like the firm.
  • A site-note protocol. How to turn a voice memo into a work-order draft without putting a homeowner’s name into a chat.

The curriculum arc

Week 1, map the week and the forbidden fields. Week 2, quote frames. Week 3, knowledge base pages. Week 4, vendor mail. Week 5, site notes and de-identification. Week 6, review with a superintendent who had not attended every session.

What changed in the work

The desk could produce a first draft while the estimator was still on the highway. The draft was visibly incomplete in the right places — blank quantities — which is a feature. New office staff had pages to read instead of “watch how I do it once.”

What stayed hard

Change orders that mix three trades still needed a person who had walked the job. The assistant could not see the site. When someone tried to let it “just fill in similar jobs,” the similar jobs were not similar. We wrote that failure into the runbook so the next hire would not have to discover it in front of a client.

Program at a glance

  • Format — morning stream, Eastern Time
  • Length — 6 weeks
  • Group — office and site leads, up to 14
  • Artefacts handed over — quote frames, assembly pages, vendor set, site-note protocol

Build 3

A community college department

Sector — teaching · Format — instructor cohort · Length — 3 weeks plus a policy workshop

The situation

Faculty were already using public assistants for outlines and example problems. Students were using them too. The department had a paragraph in the syllabus that satisfied nobody: too vague to guide a marker, too stern to match what instructors were doing at 11 p.m. on a Sunday.

The work was not to police a generation. It was to make the department’s own use visible, to redesign a few assessments so process showed, and to give instructors language for the first five minutes of class.

What we built

  • A materials checklist. Source, date, and a human pass for errors a model loves (fake papers, wrong regulation numbers).
  • Assessment notes. Where an oral checkpoint or a lab book still earned its keep.
  • A classroom script. Twenty minutes on what the tool does well and a live failure, on purpose.
  • A draft honesty statement. For the course outline, written so students could tell what must be their own reasoning.

The curriculum arc

Week 1, instructors’ own workflow and disclosure. Week 2, assessments and the live-failure demo. Week 3, outline language and a marker’s note. Workshop: alignment with the college’s academic-integrity office — they remain the authority; we supplied a working draft.

What changed in the work

The department could point to a shared stance instead of fourteen private ones. Instructors who had been hiding their own use could describe it in the same sentence they used for students. A few assignments got slower and better: more process, less take-home prose that a chatbot could finish in one pass.

What stayed hard

Detection tools remain a weak story. We did not sell them. Large classes still cannot interview every learner. The honest limit is design plus conversation, not software that claims to know. We said that in the room, including to people who wanted a product that would “just catch it.”

Program at a glance

  • Format — evenings, live online
  • Length — 3 weeks + workshop
  • Group — a department teaching team, up to 14
  • Artefacts handed over — checklist, assessment notes, classroom script, outline draft

Build 4

A bilingual nonprofit

Sector — community organization · Format — evening cohort · Length — 4 weeks

The situation

Donor reports, volunteer onboarding, and a public inbox arrived in English and French. The staff bench was thin. People pasted paragraphs into a consumer translator-chat hybrid and hoped. Some donor sentences came back fluent and slightly wrong on the program’s actual constraints. Volunteer pages contradicted each other depending on who last edited them.

We do not teach in French yet. Delivery was English, with a working rule for how to use translation tools on text the organization already owned, and how not to put a client story into a public model to “just polish it.”

What we built

  • Donor-report frames. What a model may reshape (clarity) and what it may not (outcomes, quotes, numbers).
  • A translation protocol. Human review for public French or English; no client narratives in consumer tools.
  • A volunteer knowledge base. The ten questions that generate the most mail, written once, dated, bilingual by human translators the organization already used.
  • A privacy card for stories. How to de-identify an example before anyone asks a tool to help with tone.

The curriculum arc

Week 1, inventory of texts and red lines. Week 2, donor frames. Week 3, volunteer pages and the translation protocol. Week 4, review of a de-identified report pack and a mail week.

What changed in the work

The public pages stopped drifting. New volunteers could be pointed at a dated set instead of a chain of forwarded emails. Drafting a donor paragraph became a two-step: structure from the frame, numbers from the person who held the spreadsheet.

What stayed hard

True bilingual quality still needs a person. A model will happily produce French that a native reader finds off, or English that flattens a community’s own wording. We treated machine translation as a sketch, never as a publish button, and we did not pretend our English-language classroom replaced a translator.

Bring us your own shape.

If your week looks like one of these rooms — or like none of them — write to us with the role, the group size, and the constraint you cannot drop. We will tell you whether a cohort, a studio, or a two-day intensive is the honest fit.

Book a call See the courses