Episode 166 · Green, until it wasn't

Leading people and stakeholders

Thursday, 17:05. Five weekly updates in a row said "Green, on track", with tasks done climbing to 81%. Then Atlantica's CTO asks Karim, on a call, whether the pilot is slipping three weeks. Karim didn't hear it from you.

Agents write the code, but people fund it, wait for it and decide about it. Tell them what matters, early, and give them choices.

30 min advancedStatus updates that say what matters (outcome, risks, decisions)Running a risk register (likelihood, impact, owner, trigger)Explaining trade-offs in each stakeholder's termsSaying no with reasons and an alternativeOnboarding people into an AI-first workflowA good handoff to a human expert

Problem
Week 6 of StayHub for Chains, six weeks before Atlantica's 20-hotel pilot. Five weekly updates said "Green, on track" with a rising percentage of tasks done. Karim learns from Atlantica's CTO, on a call, that the pilot may slip three weeks.
Why the simple system fails
Reporting activity (tickets closed, agent PRs merged) instead of outcomes and risks. With agents, activity numbers grow fast whether or not the project is on track, so they hide exactly what stakeholders need: what might go wrong, and what they must decide.
Engineering concept
Status = outcome and confidence, what changed, top risks with owners and triggers, and decisions needed by whom and when. A risk register keeps risks visible early. Trade-offs are explained in the listener's terms; a no comes with reasons and an alternative. People joining get the same rules agents get, enforced.
Technology
A weekly written update, RAG status with a reason, a risk register, hill charts, decision requests with options and costs, an onboarding checklist (CLAUDE.md, permissions, required checks, pairing), an expert-handoff brief.
Trade-offs
Honest early bad news costs comfort and sometimes credibility in the moment; late bad news costs trust and options. More detail isn't more clarity. Bringing in an expert costs money and time; not bringing one in can cost a one-way door.

The problem

2 min

Thursday, 17:05, week 6. The pilot (20 Atlantica hotels) is due in week 12.

Karim · CTO
Atlantica's CTO asked me if the pilot is slipping three weeks. Their security team is worried about single sign-on.
Karim · CTO
Your last five updates said Green. Tasks done went from 52% to 81%.
You · Engineer at StayHub
All true. SSO has been stuck since week 4, but everything else is moving.

The big idea

4 min

81% done, three weeks late

The big idea

With agents writing the code, your job moves to people: report outcomes and risks early, turn problems into decisions, and say no with an alternative.

A pilot doesn't announce "we've used 68% of our fuel". They say "we'll land 20 minutes late; connections after 15:00 are safe", and tell the tower something else, in its terms, with a request.

How it works

5 min

A is green outside and red inside: activity reported until the surprise. Each needs something different: Karim the date confidence and decisions in cost and risk; Maya what Atlantica gets and what's cut; Atlantica dates they can trust.

The weekly update, and what feeds it

1
Outcome + colour
A colour (red, amber, green) is a forecast with a reason: "20 hotels in week 12? Amber, 60%, because SSO is stuck."
2
What's uphill
Place each workstream on a hill: uphill is still figuring it out, downhill is known execution. SSO has sat uphill since week 4.
3
Top risks, from the register
A lists each risk with owner, trigger and response, reviewed weekly. Go amber when a trigger fires: cheap options expire first.
4
Decisions needed
One person, a date, priced options, your recommendation: "Karim, by Monday: expert (€18k) or a weaker pilot login. I recommend the expert."
5
No, with an alternative
Reasons, an alternative that gives most of the value, and the price of yes. That makes refusal a choice they can make.
6
Newcomers and experts
Project agent settings committed to the repo, no admin bypass, a buddy in week 1. When nobody can judge an answer, bring in an expert, not a better prompt.

Watch a decision being made

4 min

Worked example

You rewrite the week 6 update; Lina only checks

Turn week 6's facts into an update Karim and Maya can decide from in five minutes, sent before Friday.

1Outcome and an honest colour
Amber, 60%, because SSO is stuck. Not red: options still save the date. ARI push and invoicing are downhill; SSO uphill since week 4: nobody has shipped SAML, and nobody can judge the agent's login.
Why: Amber means "at risk, with a plan"; red means "off track, decide now".
2Top risks with owners and triggers
SSO (you; no expert-reviewed design by week 8), review capacity (team leads; oldest PR > 2 days), CRS vendor (Maya; < 20 hotels pushing by week 9).
Why: Triggers agreed in advance end arguments about whether it's "really" a problem.
3Turn the problem into a decision
Karim, by Monday: (a) Hana, identity specialist, 3 weeks, ~€18k; date holds. (b) StayHub logins with 2-step verification for the pilot, SSO by week 20, if Atlantica's security accepts. I recommend (a).
Why: A bare "SSO is at risk" would have triggered anxiety and a meeting.
4Cut the activity
PR counts and percentages go behind a link. Lina: "Can Karim decide from this without asking a question?" Yes. He chose Hana on Friday; she found the login accepted assertions meant for another service, and replayed ones.
Why: Every line that doesn't help someone decide hides the lines that do.

In real life

3 min

Count the tickets, or show the unknowns?

Burndown / % complete (Jira, Scrum tools)

Tickets done vs planned
Easy to generate
Blind to unknowns

Shows activity; hides where work is stuck.

Hill charts (Basecamp, Shape Up)

Uphill: still figuring it out
Downhill: known execution
Published as a free book

Shows uncertainty; needs honest judgment.

What goes wrong

3 min

Waiting to be sure before going amber

Cheap options expire first.

Instead

Amber when a trigger fires, with a plan.

A problem without options, or a bare no

Stakeholders can't act; they escalate around you.

Instead

Priced options and a recommendation; a no with an alternative and the price of yes.

Newcomers on personal setups with override rights

One bypass is enough to hit shared tables.

Instead

Project settings in the repo, no admin bypass, a buddy in week 1.

Trade-offs

3 min

Approach	Fits when	You pay
Report early and plainly	Almost always	Harder talks sooner; some ambers turn out fine
Report when certain	Small risks the team absorbs alone	Options expire; one surprise taints every green
Say yes, add agents	Free review capacity, two-way doors	Review bottleneck; risk grows faster than output

Week 7, Monday, 10:00. Atlantica's revenue director wants a chain-wide promotions editor.

Maya · Product manager
I'd like it in the pilot. It's what their revenue team will judge us on.
Karim · CTO
The date doesn't move. Run four more Claude Code sessions and do both.
Maya · Product manager
Or move the pilot three weeks and do it properly.

Your job as tech lead

4 min

Claude Code can write the code for this. You still own the decisions, and you have to know what good looks like to accept or reject its work.

Decisions you own

The status colour and your confidence in the date: a forecast you own, never generated from ticket counts.
Which risks go in the register, their owners and the triggers that start the response.
What you ask each stakeholder to decide, what you say no to, and the rules newcomers follow.

Tell Claude Code

Draft this week's status update for StayHub for Chains from docs/status/week-06-notes.md, docs/risk-register.md and docs/plan.md. Five parts: outcome and colour, what changed, top three risks with owner and trigger, decisions needed (who, by when, priced options), a link for detail. Don't pick the colour or recommendation: list evidence for each. No PR counts or percentages. Mark anything you couldn't verify.

Check before you accept

You chose the colour; it leads, with a reason.
Each decision request names one person, a date, priced options and your recommendation.
Read it as Karim: can you decide without asking a question?

Red flags

Task completion turned into a colour ("81% done, so green").
"Some challenges", "on it": risk wording with no owner, trigger or number.
Facts that aren't in any source file: a plausible guess filling a gap.

Check yourself

2 min

Q1/3Scenario

Which status line is most useful to Karim?

How confident are you?

Next episode, the last: a new quarter, a new project, and this time Lina won't explain anything. You lead it from the goal to the retrospective.

Finished?

Mark it complete to update your skill map.

Next: Capstone: lead a quarter-long project

Go deeper (optional)You don't need this to understand the lesson. Open it when a project needs the details: Technical deep dive, Analogy, Why it matters, Check your understanding, Practice ladder, Explain it yourself.

Technical deep dive

Rewriting the week 6 update

Worked example

Rewriting the week 6 update (Lina only checks)

Turn the facts of week 6 into an update Karim and Maya can decide from in five minutes, and send it before Friday.

1Outcome and an honest colour
Goal: 20 Atlantica hotels live in week 12, rate changes in search within 5 min. Colour: amber, 60% confidence, because SSO is stuck. Not red: there are options that still save the date.
Why: The colour is a forecast with a reason, not a mood. Amber means "at risk, with a plan"; red means "off track, needs a decision now".
2Find what's uphill
Place each workstream on a hill: ARI push downhill (batch applier done, 2.5 min p95 in staging); chain invoicing downhill (golden tests nearly built, seam next week); SSO uphill since week 4 (Atlantica requires SAML; nobody on the team has shipped it; the agent's login works but nobody can judge its security).
Why: Work stuck uphill is where dates slip. Two weeks uphill was the signal the old update hid.
3Pull the top risks from the register
Three rows, each with owner and trigger: SSO (you; no expert-reviewed design by week 8), review capacity (team leads; oldest PR waiting > 2 days), Atlantica's CRS vendor (Maya with Atlantica; < 20 hotels pushing in sandbox by week 9).
Why: Triggers agreed in advance mean nobody argues later about whether it's "really" a problem.
4Turn the problem into a decision
Karim, by Monday: (a) Hana, a freelance identity specialist, 3 weeks, ~€18k; date holds. (b) Ask Atlantica to accept StayHub logins with 2-step verification for the pilot; SSO by week 20; depends on their security team. I recommend (a).
Why: Options with costs, a recommendation and a deadline let him decide in five minutes. A bare "SSO is at risk" would have made him anxious and called a meeting.
5Frame it for each reader
For Karim: €18k against three weeks and Atlantica's trust. For Maya: what Atlantica's staff see on day one, and that the promotions editor stays out of the pilot (a separate note with reasons and an alternative).
Why: Same facts, different terms. Karim decides money and risk; Maya decides scope and what the customer is told.
6Cut the activity
PR counts, ticket percentages and velocity go behind a link to the board. Lina's only check: "Can Karim decide from this without asking you a question?" Yes.
Why: Detail isn't clarity. Every line that doesn't help someone decide makes the important lines harder to find.

Artefact	Who reads it	What it must let them do
Weekly update	Karim, Maya, Atlantica's lead	Decide: fund, cut, re-date, tell customers
Risk register	You, squad leads, Karim	Act early when a trigger fires
Decision request	One named person	Choose between priced options by a date
"No" with alternative	The requester (Maya)	Get most of the value without breaking the plan
Onboarding checklist	Newcomers and their buddy	Work in the AI-first workflow without skipping rules
Expert handoff brief	Hana	Start useful work on day one

Analogy

A pilot on a long flightA tech lead on a big project

A pilot on a long flight"We'll land 20 min late"A tech lead on a big projectOutcome and honest colour
A pilot on a long flightNot "68% of fuel used"A tech lead on a big projectNot "81% of tasks done"
A pilot on a long flightDifferent words for the towerA tech lead on a big projectDifferent framing per stakeholder
A pilot on a long flight"Not that runway; this one"A tech lead on a big projectNo, with an alternative
A pilot on a long flightCalls the airline engineerA tech lead on a big projectBrings in an expert

Where it breaks: passengers can't change the flight plan. Your stakeholders can, which is exactly why they need to hear early.

Why it matters

Early news or late news?

Amber in week 4

Hire an expert
Cut a feature
Agree a new date

Three options, all still cheap.

Surprise in week 6

Customer already planned
One option gone
Trust takes a hit

Fewer options, more meetings.

Bad news only gets more expensive with time. People can only fund an expert or move a date if they know.

And trust is the real budget. After one surprise, every green gets questioned. You end up reporting instead of leading.

Check your understanding

Q1/3Reasoning

Put the parts of a good "no" in order.

1.The reasons, in the listener's terms
2.What it would take to say yes
3.The answer, plainly: not in this release
4.What we can do instead

How confident are you?

Practice ladder

Step 1 · Trace

What will they learn too late?

Scenario Warm-up

Squad lead's update to you, week 8:

"Green. ARI push: 23 PRs merged, all checks green, 97% of tickets done. Atlantica sandbox: working on it with their vendor. Load test postponed to next sprint because the environment wasn't ready. Two contractors now fully autonomous."

Trace what each line hides. For each, say what you'd ask, and which stakeholder would be surprised later if you passed it on as is.

Step 2 · Complete

Complete the risk register

Design Practice

Fill the blanks. A trigger must be observable and early enough to act.

Risk	Owner	Trigger	Response
SSO not secure or ready	You	___	___
Review capacity overwhelmed	Squad leads	___	___
Atlantica's CRS vendor late	Maya (with Atlantica)	___	___
Chain invoices diverge from ledger	Inès	___	___

Step 3 · Solve

Say no to Karim

Explain Stretch

Week 9. Karim: "Nordlys Hotels (60 hotels in Norway and Denmark) want in. Add them to the week 12 pilot. Same code, more hotels, right?"

Facts: Nordlys uses a different CRS vendor whose push format nobody has tested; Norway's currency (NOK) and Denmark's (DKK) haven't appeared in chain invoicing samples; review capacity is at its limit; Atlantica's pilot is the contract that matters this quarter.

Write your reply to Karim: the answer, the reasons in his terms (cost, risk, date), an alternative, and the price of yes.

Step 4 · Decide

Expert, agent or service?

Decision Stretch

Atlantica's Italian hotels, like every other Italian partner hotel, must receive StayHub's invoices through Italy's national e-invoicing system (a government exchange with a strict XML format and legal rules). Nobody at StayHub has done it. Options:

Let Claude Code read the official specification and build it; the team reviews.
Bring in an Italian e-invoicing consultant for two weeks (~€9k) to design it with the team, then agents build it.
Use a certified e-invoicing provider that submits invoices for you (~€0.30 per invoice, ~1,500 invoices a month for Italy).

Write your decision: "We chose X over Y because…, accepting the risk of…".

Explain it yourself

A new tech lead asks: "If the agents do the coding, why do you spend so much time on status updates and meetings?" Answer in a few sentences.

0 words