This is the full reference version of Session 2 of the AI Bootcamp: what changed in Cowork, context files, skills on a schedule, and how to keep AI work off the slop pile.
Switch back to the presentation view with the button in the corner, or press E. Use M for the menu at any time.
Make it yours: a smarter Cowork, context files, skills on a schedule, and the quality bar.
Me before AI Bootcamp: 
Me after AI Bootcamp: 
Accurate. But handsome is not a one-time upgrade, it is a habit.
Today: how to stay Handsome Squidward when the novelty wears off.
Four parts, three live demos, and one challenge you can finish by Friday.
What changed since Session 1, and how to push through a bad first draft
Stop re-introducing yourself in every chat
Teach it once, put it on a schedule, let it come to you
The quality bar before anything leaves your desk
The tool you learned in Session 1 has been quietly getting stronger. Four upgrades change how you should work with it, starting today.
Scheduled tasks no longer need your laptop open or your PC on.
Two modes, two jobs. Knowing when to flip the toggle is half the skill.
Subagents, models, and effort: more thinking power when it matters.
What to do when the first output is bad. Hint: not giving up.
The biggest change since Session 1: scheduled tasks now run in the cloud. Close the laptop, shut down the PC, go on vacation. The work still happens.
Your scheduled task was tied to your machine. Laptop closed at 7am? The 7am task never ran. The automation was only as reliable as your charging cable.
Tasks run in the cloud on their schedule, whether your machine is awake, asleep, or in a bag. Results are waiting for you when you show up.
Same brain, two working styles. The toggle at the top is the most underused control in the product.
A conversation. You stay in the loop.
Quick questions, summaries, and explanations.
Drafting and editing where you want to steer each round.
Thinking out loud: pressure-testing an idea before a meeting.
A delegation. It works, you review.
Multi-step jobs: research, build, and assemble in one run.
Anything touching files: documents, spreadsheets, decks, folders.
Long tasks you would rather not babysit. Set it going, come back.
Rule of thumb: if you would forward it to a capable assistant, it belongs in Cowork.
Default settings are tuned for everyday work. When the problem is genuinely hard, you can turn up the machine. All three dials exist in Claude Cowork and Claude Code.
Parallel workers on one job. One researches, one drafts, one checks. Big tasks stop being serial.
Pick the heavyweight for the hard call, the fast one for the routine sweep. Match the brain to the problem.
Let it think longer before it answers. Cheap insurance on decisions you will live with for months.
It will happen. What you do next is the whole game.

Gives up. Or worse, ships it.
Tries once, gets a mediocre result, declares AI overrated.
Never says what was actually wrong with the output.
Starts from zero next time and gets the same mediocre result.

Treats it like feedback, not failure.
Tells it exactly what is wrong, in plain language.
Has it write the lesson into its memory or its skill.
Runs it again. The next version, and every version after, is better.
We build something real, it comes out rough, and we fix it in front of you. No edited highlights: this is the actual loop you will use every day.
What to watch for
Every great prompt you have ever written had one thing in common: context. A context file means you write that context once, and every conversation starts warm.
Why pasting the same background into every chat is costing you the output quality you want.
A document the AI reads automatically: your role, your standards, your world.
We create one for writing style from real samples, and watch the output start sounding like you.
A context file is a document the AI reads at the start of every conversation. Write the background once, never type it again.

Re-introduces himself every single chat.
Types 'I'm a customer success manager at a payroll company...' for the 40th time.
Forgets half the background, gets generic output, blames the AI.
Every conversation starts cold.

Wrote it down once.
Role, team, product, audience, and standards live in a context file.
Every chat starts already knowing his world.
His prompts are one sentence long and his outputs are specific.
We feed it two or three things I actually wrote, have it extract the style, then generate something new. The goal: output that sounds like me, not like AI.
What to watch for
Writing style is just the opener. Each of these takes 15 minutes to build and pays you back on every conversation after.
What you own, who you work with, what good looks like in your job. Suddenly its suggestions fit your actual responsibilities.
Features, jargon, pricing, roadmap. It stops explaining your own product back to you incorrectly.
Who they are, what they worry about, how they talk. Everything you draft lands closer to what they need to hear.
Pick one. Build it before Friday. That is half of today's homework already done.
A context file tells the AI who you are. A skill tells it how you do a specific job. Put a skill on a schedule and you have built your first real automation, without writing a line of code.
Meeting notes in, tight recap out: owners, dates, house style. Built from one good example.
The skill runs on a cadence you set, and per Part 1, your laptop can stay closed.
The output shows up ready for review. You went from doing the work to approving it.
Meeting notes go in, a tight recap comes out: decisions, owners, due dates, in house style. We teach it once, then rerun it on fresh notes to prove it stuck.
What to watch for
A skill you have to remember to run is still a chore. A skill on a schedule is an employee.
Build the skill once, from one good example. Correct it until the output is right.
Make it a recurring task in Cowork: every Friday at 4pm, every morning at 7.
It runs in the cloud and the result lands in front of you, ready to review and send.
That is a personal automation. No ticket, no engineer, no budget line.
AI made content nearly free to produce. It did not make it free to read. The last part of this session is about the quality bar: what has to happen before AI output leaves your desk.
Unreviewed AI output shipped to a coworker as if it were finished work.
Easy to generate, expensive to consume. Someone always pays; the question is who.
Trim it, add your judgment, and never let it look like AI wrote it.
Same tools, same hour of effort. Completely different reputation six months from now.

Ships AI slop.
Generates 10 pages, skims none of them, hits send.
Makes his coworkers do the reviewing he skipped.
Everyone quietly starts ignoring what he sends.

Instructs, plans, then inspects and verifies.
Briefs the AI properly, with context and a plan.
Inspects the output line by line and checks the claims.
Ships two tight pages he would put his name on. Because he did.
The math of shipping a raw AI document to your team, in round numbers:
30 min
You saved by not editing the 10 pages down.
5 x 20 min
Five readers each spend 20 minutes wading through it to find the four sentences that matter.
Ten pages can almost always be two. Cut it down to what the reader actually needs, then cut again.
Your read on the client, the political context, the call you would make. That is the part AI cannot do, and the part your name is for.
Kill the telltale headers, the bullet avalanches, the 'delve' and 'moreover'. If it reads like a bot, it gets treated like spam.
What is in the way of you using this tomorrow morning?
Not a study plan. Two small builds, both demoed today, both under 20 minutes.
Your writing style, your role, or your product. Pick the one you repeat most often.
Teach it one job you do weekly, put it on a cadence, and let it run with your laptop closed.

Stuck? Post in your internal AI channel.
Want more? A hands-on power user session is coming.