Imaguru x Projector Community

AI for managers

Follow-ups, status reports, idea-to-prototype. How AI fits into a real working day.
Wednesday, 30 September 2026 / 18:30 CEST / Imaguru, Warsaw
Andrew Tomin: AISandboxLab.com
Senior Product Manager
Founder, AI Sandbox
Andrew Tomin presenting the anatomy of an AI agent at Imaguru
AI for Managers at Imaguru, Warsaw
1
No laptop needed.
Tonight you watch and check.
2
Take the prompt pack home:
aisandboxlab.com/imaguru-managers
3
Think of one thing you wrote this week that you'd rather not write again.
QR code: aisandboxlab.com/imaguru-managers
Prompt pack
01
Honest conversation
18:45

Where does AI create friction in your work today?
Where could it genuinely help?

Friction
Could help
02
Before the demos
18:50

Chat and agent workflows

A chat answers once. An agent works through a task, uses project files and checks the result before handing it back.

Chat You paste text in AI writes one answer You copy it out One question, one answer. You carry everything between steps. Agent Your goal + your files 2. Act read, write, build 3. Look at result 4. Fix or go on 1. Plan the steps repeats until the goal is done Result files you open You check then send
03
Before the demos
18:51

Anatomy of an agent

The model sits inside a larger system of context, rules, tools and checks.

Context CLAUDE.md, project files, data Skills and rules SKILL.md: how to do the job Memory what it knows between sessions Model: the core (Claude) thinks and decides the next step Tools read, write, search, run commands Connectors (MCP) email, Jira, databases, other services Checks tests, sources, visible results Browser open a site, check it, click Where the agent acts files · browser · external services · repositories The agent loop: Understand Plan Do Check Repeat
04
The frame for every demo
18:52

A working frame for every AI task

01
Input
Real material: a transcript, updates, an export, three lines of an idea
02
Agent steps
What it does, in what order. The prompt says this for you
03
Human check
What could be wrong, and how you spot it. The one people skip
04
Output
Something a team can actually use: a message, a report, a link to click
05
Tonight's story
18:53

One week at Kestrel Freight

A fictional B2B delivery company in Warsaw. Customer Portal release 2.3. All people and numbers are invented.

Monday 21 Sep

A messy weekly sync

Six minutes, two release dates, one task nobody owns.

Wednesday 23 Sep

Anna, the COO, wants a status

Three days of chat and a Jira export. Nothing a COO can read.

6 October

Planning session

140 reschedule calls a month. Marta promised something to click.

Marta, Product ManagerTomasz, Tech LeadAisha, QA LeadDaniel, Customer SuccessKasia, Delivery Manager
06
18:55 / 12 MIN
Demo 1

Meeting transcript to decisions, owners and follow-ups

Input
Monday's sync, auto-transcribed
Agent steps
Decisions, owners, dates, open questions, draft message
Human check
You, with the room
Output
Reviewed follow-ups, ready to send
07
Demo 1 / Human check
Before you send

Human check: meeting follow-up

  • Is every decision the final one? Meetings change their mind halfway.
  • Does every task have an owner who actually said yes? "We'll sort it" is not an owner.
  • Are any dates made up? "Early next week" is not a date.
  • Is there anything in the output that nobody said? Check the source column.
  • Would each owner recognise their task if they read it cold?
08
19:07 / 12 MIN
Demo 2

Daily updates and Jira to a status report for the COO

Input
3 days of team chat, 15 Jira tickets
Agent steps
Colour, changes, risks, asks, conflicts. Two versions
Human check
Is the colour right?
Output
One page Anna reads in two minutes
09
Demo 2 / Human check
Before you send

Human check: status report

  • Is the status colour right? Would you defend it to the COO?
  • Does the tracker say "Done" for something a person says is still broken? Trust the person, then fix the tracker.
  • Is there a risk that lives only in chat? It still belongs in the report.
  • Does every risk have an owner and a next step?
  • Is the ask something the stakeholder can say yes or no to?
10
19:19 / 14 MIN
Demo 3

Idea to spec and clickable prototype

Customers who can't be home for their delivery slot call us to reschedule. 140 such calls last month, up from about 90 in June, about 8 minutes each.

Idea: a "Can't make it? Pick another slot" link in the delivery SMS.

Time slot only, no address change. Should look like a real app, with the delivery on a map. Clickable for planning on 6 October, not connected to real dispatch data.

11
Demo 3 / Human check
Click it yourself

Prototype acceptance checks

  • 1The SMS link opens the delivery, the current slot and a map
  • 2Full slots are marked and cannot be picked
  • 3Continue stays disabled until a free slot is picked
  • 4Old and new slot side by side before confirming
  • 5Confirmation shows the new slot and promises an SMS
Output

A conversation piece for the planning session. Not the product, not connected to anything real. Click it on your phone: aisandboxlab.com/imaguru-managers-demo

QR code: aisandboxlab.com/imaguru-managers-demo
Live prototype: aisandboxlab.com/imaguru-managers-demo
12
Where this breaks, and your Monday
19:33

Three rules

  • 1Never paste what you would not email to a stranger.Client data only in a tool your company approved for it
  • 2The AI drafts, you sign.Tonight the check caught a changed date, a missing owner, a wrong colour
  • 3Start with one real recurring task.The follow-up or status report you write every week
13
Questions
19:35

Your questions

And one line of feedback, please: scan, and Telegram opens a message to me. It tells me what to show next time.

Audience at the AI for Managers workshop at Imaguru
QR code: Telegram message to Andrew Tomin
Feedback on Telegram
14
Facilitated networking
19:45-20:15

Networking discussion

Table 1

Follow-ups

Table 2

Status reports

Table 3

Prototypes

Round 1 / 19:47

One recurring task you'd hand to AI first. What would your human check be?

Round 2 / 19:59, switch tables

What stops you from trying this next week? Data rules, tools, time or habit. One way around it.

15
Thank you
20:15

One task for next week

Prompt packaisandboxlab.com/imaguru-managers
Prototypeaisandboxlab.com/imaguru-managers-demo
Siteaisandboxlab.com
LinkedInlinkedin.com/in/andrew-tomin-senior-product-manager
Telegramt.me/AndrewTomin
QR code: aisandboxlab.com/imaguru-managers
Prompt pack
16
Speaker notes
Arrows / Space: navigate  ·  N: notes  ·  P: print mode  ·  F: fullscreen