Tooling & Onboarding
The one prerequisite for both Innovation Day sessions is simple: bring at least one working AI chatbot you can log into. This doc is how we make sure that’s actually true before the room goes quiet at 09:00 - and what we do for the people for whom it isn’t.
Everything here is tool-agnostic. ChatGPT, Gemini, Claude, Copilot Chat - all fine. We never assume a specific tool. We only assume you can sign in and send one message.
The mixed estate is the point, not the problem. ~70% of attendees have one tool; the rest are on OpenAI or Gemini; Madrid HR/Procurement are mostly Gemini; and not everyone has a licence. Both sessions were designed to run on any of these. Nobody is disadvantaged by which tool they bring. See the five ways of working for the techniques - they’re identical across tools.
The golden rule (applies to every path below)
Synthetic or public data only. No real names, no client data, no PII, no confidential figures - on any tool, including a fallback or a shared login. Anonymise first: real names become “Person A”, vendors become “Vendor 1”. When in doubt, leave it out. This is Way #3: Mind what you feed it, and it is non-negotiable on every path in this document - especially the loaner and shared-login fallbacks, where you are typing into an account that isn’t yours.
BEFORE the day - confirm you can log in (2 minutes)
Do this once, any time before Monday. It is the single highest-value thing you can do to not lose your first 20 minutes.
-
Open the AI tool you put on your sign-up form. The exact one you’ll bring.
-
Log in. Not “I’m pretty sure I have a licence” - actually reach a working chat box.
-
Run the smoke test. Paste this one line and check you get a sensible answer back:
In one sentence, what is spec-first development? -
You pass if you got a coherent one-sentence answer. (Doesn’t need to be perfect - you’re testing the login and the pipe, not the model.)
If any step fails - login loops, “no licence”, a blank screen, a corporate block - reply to [ORGANISER NAME / CHANNEL] straight away. Telling us on Friday is a five-minute fix. Telling us at 09:00 is a fallback. Don’t sit on it.
VERIFY - how we make sure the room is ready
We don’t rely on hope. Three checkpoints, escalating only as needed:
| When | Check | What it catches |
|---|---|---|
| At sign-up | Form questions 5 & 6 - which tool and can you log in today (yes/no) | Tool-less people and unactivated licences, surfaced before rosters freeze |
| T-2 days | Targeted reminder to anyone who answered “no” / “I don’t have one”, plus a one-line nudge to everyone to run the 2-min smoke test | People who meant to check and didn’t |
| On the day (T-0) | 5-minute door check, folded into the session opener | The last stragglers - handled live, no separate session |
The door check (folded into the opener)
The first five minutes of each session are the verification - no separate setup slot. While the facilitator frames the five ways of working, the co-facilitator runs the room:
“Everyone open your AI tool now and send it: In one sentence, what is spec-first development? Thumbs up when you’ve got an answer back.”
Anyone without a thumbs-up gets routed straight into the fallback ladder below while the opener continues. No one is singled out, nobody waits, and the session starts on time.
On the day - the FALLBACK LADDER
If someone can’t get a working chatbot, work down this ladder in order. Stop at the first rung that works. Most people land on rung 1 or 2.
Rung 1 - Pre-vetted no-login free tier
Have one browser tab bookmarked to a mainstream chatbot’s free, no-login tier that we’ve tested on the venue network beforehand. Hand the person the laptop or the URL; they’re chatting in under a minute. No account, no licence, no approval.
- Why first: zero friction, no credentials, no waiting on anyone.
- Prep: the facilitator confirms the chosen free-tier URL loads on the room Wi-Fi during the network pre-check - including in Madrid.
Rung 2 - Pair the tool-less person as spec-writer / reviewer
Pair them with someone who does have a working tool. The tool-less person drives the thinking; their partner drives the keyboard.
- Build a Skill: they write the Goal, Inputs (with the PRIVACY CHECK), and Output sections - the spec-first half - while their partner runs the prompt.
- Spec-First Build: they take the spec-writer / reviewer seat - authoring the contract and running the human review gate (Way #2), which is the role that needs a person, not a licence.
- Why this is good, not a consolation prize: it’s pedagogically the strongest seat in the room. Way #1 (describe before you generate) and Way #2 (a human always signs the work) are exactly what this person practises, undistracted by a chat window. Frame it that way out loud.
Rung 3 - Supervised loaner / shared login
Last resort. A facilitator provides a loaner laptop or a shared login on a tool we control, used under supervision.
- Conditions: facilitator-supervised, signed out at the end, and - like everywhere else - synthetic/public data only. A shared account is more sensitive about data, not less.
- Why last: credentials and accountability are muddier than rungs 1–2, so we only reach for it when pairing genuinely isn’t possible.
| Rung | Path | Reach for it when |
|---|---|---|
| 1 | Pre-vetted no-login free-tier tab | Always try first - fastest, no credentials |
| 2 | Pair as spec-writer / reviewer | They’d benefit from the thinking seat, or rung 1 is unavailable |
| 3 | Supervised loaner / shared login | Pairing isn’t possible and they need their own hands on a tool |
The silent killer - test the network FIRST
The thing most likely to sink the session is not licences. It’s the venue Wi-Fi or corporate proxy quietly blocking the AI domains. A room full of valid logins is worthless if the network drops the requests. This has bitten us before - treat it as a hard pre-check, not a nice-to-have.
Before the day, from the actual room (and on the actual guest/corporate network attendees will use):
- Load each mainstream chatbot you might rely on - ChatGPT, Gemini, Claude - and send one real message on each. A homepage that loads but a request that hangs is the classic failure.
- Test the rung-1 no-login free-tier URL specifically - that’s your safety net; it must work.
- Test on both the guest Wi-Fi and the corporate/proxied network, since attendees may be on either.
- Run this check in the Madrid office explicitly. Different proxy, different filtering rules, mostly Gemini users - do not assume the HQ result carries over. Test Gemini there in particular.
- If anything is blocked, escalate to IT/networking before Monday with the specific domains, or arrange a mobile hotspot / MiFi as backup for the room.
If the network can’t be fixed in time: a couple of mobile hotspots covering the room beats fifteen people staring at a spinner. Know this before the day, not during the opener.
Onboarding done right means the session starts on the five ways of working, not on IT support. Get the logins and the network sorted before Monday and the room runs itself.