A chart's plot area: a SlideCard of fixed height the chart fills.

claude --print "fix the failing test" # one run

neutral

A group

primary

A group

primary, active

A group

secondary

A group

secondary, active

A group

size xl

Ship the loop

size lg

Ship the loop

size md

Ship the loop

size sm

Ship the loop

tone solid

33310x

tone soft

33310x

tone badge

33310x

tone ink

33310x

The caption under a diagram, its last step.

neutral
quiet
primary · tint
primary · soft
primary · outline
primary · line
secondary · tint
secondary · soft
secondary · outline
secondary · line
accent · tint
accent · soft
accent · outline
accent · line
success · tint
success · soft
success · outline
success · line
error · tint
error · soft
error · outline
error · line

The line under a headline, in muted ink.

A slide's headline on one line

The column of a titled slide: headline, then the body under the fit guard.

The Factory
New stage
Slide Catalogue

Every slide type and its variants, with fictional content

Primary themeOne loop per team

Content slide with kicker, principle pill and footer

A subtitle states the slide's claim in one sentence.

A paragraph carries one thought with an optional link.

  • Short bullets read faster
  • Nested items
    • One level

Footer line: the take-away the speaker repeats.

007 / 105

Centered layout

A centered hero for a question or a demo moment

Centered content blocks over the particle background.

One line under the hero.

013 / 105

Regular slide with a profile sidebar

Mara Lindqvist

Mara Lindqvist

Platform lead, Lumen Freight

What she built

  • One harness
  • Shared checks

Why

Teams kept solving the same problem.

Read more
043 / 105

Regular slide, cards layout

  1. 1.FoundationsTools and environment
  2. 2.DeliveryPlan, build, review
  3. 3.LearningConventions and metrics
044 / 105

Code first, plan later

Plans before code

Review reads the plan before anyone writes a line.

01The plan is reviewed
02Code follows the plan
03Review reads both

That is the whole method.

046 / 105
Grid

Four pillars

Tools

What the agent uses.

Knowledge

What it reads.

Checks

What is enforced.

Metrics

What we learn.

052 / 105

Numbers after six months

2dfrom 9dMedian, all teams.
−40%per ticketFictional.
38on the harnessOf 40.
054 / 105

What participants said

“I review plans now, not just diffs.”

Engineer, fictional

Team lead, fictional

“The checks caught what I missed.”

Team lead, fictional

“Monday metrics changed our backlog.”

Product lead, fictional

055 / 105

Prompting against a harness

Prompting

Fast to start

  • +Ad hoc
  • +Hard to review
  • +Knowledge in heads

Harness

Fast to sustain

  • +Planned
  • +Reviewed twice
  • +Knowledge in files

Pick the one you can still run next year.

Fictional comparison.

056 / 105

Before and after

Before

Slow

  • Release every month
  • Hotfix by hand
→

After

Routine

  • Release on merge
  • Rollback without a meeting

Same team, other habits.

057 / 105

Platform

  • Every pull request runs the same checks the agent ran locally, so review starts from green
  • The harness writes its own plan before it writes code, and the plan is reviewed first
  • Flaky tests are quarantined within a day and fixed by the team that owns them
  • Each team keeps one page of conventions the agent reads before every task

Teams

  • Flaky tests are quarantined within a day and fixed by the team that owns them
  • Each team keeps one page of conventions the agent reads before every task
  • A failed deployment rolls back without a person deciding it
  • Metrics are read every Monday and decide what the team stops doing
058 / 105

Writing a check the agent can run

One bad and one good version.

  • Name the command
  • Name what passes

Check

Bad

Make sure it works.

Good

Run bun run test search.
Passes when every umlaut query returns a result.
060 / 105

Two positions and a resolution

Mara Lindqvist01

Shared

“Every team should use the same harness.”

&
02Jonas Okafor

Owned

“Every team knows its own work best.”

Both

The verdict

One harness, team rules on top.

061 / 105
Media

One screenshot

Fictional board
064 / 105

Gallery of three

Board
Chart
Code
065 / 105

Row with tags

PlanBuildReview
PlanBuildReview
066 / 105

Your trainers

Mara Lindqvist

Mara Lindqvist

Platform lead

Lumen Freight

  • Built the harness
  • Ten years of platform work
Jonas Okafor

Jonas Okafor

Team lead

Lumen Freight

  • Runs the weekly review
  • Coaches new teams

Both fictional.

068 / 105
Pipeline stages
Plan
#1Ideate
#2Refine
Build
#3Implement
#4Test
Ship
#5Review
#6Deploy
074 / 105
The Factory
Foundations
Tools

What it uses

Knowledge

What it reads

Checks

What is enforced

Pipeline
Plan

Before code

Build

From the plan

Review

Plan and diff

070 / 105

From autocomplete to delivery harness

1x

Autocomplete

Line by line

Faster typing.

3x

Chat

Task by task

Copy and paste.

10x

Harness

Ticket by ticket

Plans, checks, review.

You are here

Each stage changes what a person reviews.

049 / 105
LanesIn parallel

Three tickets at once

LF-142
Plan
→
Build
→
Review
LF-143
Plan
→
Build
LF-144
Plan

Each lane is one agent.

077 / 105

Adoption cycle

Pilots15%Early teams35%We are hereMajority35%Late teams15%
079 / 105

Value over time

TodayFictional projectionTimeValue
080 / 105

Lead time fell, review wait did not

Lead timeReview wait

By September, review wait is the whole lead time. Fictional figures.

081 / 105

Build holds the most tickets

Two in five tickets sit in Build. Fictional figures.

082 / 105

Where does your team review today?

086 / 105

Part two

Review that starts from green

Why the checks run before a person looks

Mara Lindqvist

Lumen Freight

087 / 105

Thank you.

Mara Lindqvist

Mara Lindqvist

Lumen Freight

mara@example.com
Jonas Okafor

Jonas Okafor

Lumen Freight

jonas@example.com

“The best harness is the one nobody notices.”

097 / 105

Let's discuss

Which check would you add first?

Who reads the plans in your team?

What would you stop doing on Monday?

099 / 105

Roadmap

Two staged columns

What comes first, what comes after.

Part one

Ingredients

Non-negotiables

  • Plans first

    Reviewed before code

  • Shared checks

Part two

Tactics

  • Determinism

    Same input, same output

  • Learning

    Conventions grow

075 / 105

Part two · Going deeper

The operational tactics

Where each tactic sits in the pipeline.

089 / 105
PlanA month is a planning unit

The programme at a glance

M1
M2
M3
M4
M5
M6
Assessment
Baseline
Six areas
Six areas
Training
Per cohort
Cohorts
Cohorts
Mentoring
Weekly
Weekly
Meilensteine
M1: Baseline
M6: Handover
Meilensteine
Baseline
Handover

Fictional plan.

090 / 105

Four team types

Stream-aligned teams deliver, the other three make them faster.

Enabling: QualityTestingReview
Stream team: Checkout
Product leadEngineers
Stream team: Search
Product leadEngineers
Routing engineRoute optimisation
Platform teamPipelines and harness
Harness engineers

Fictional organisation.

091 / 105

Agenda

  1. 1.Where we stand
  2. 2.How the loop works
  3. 3.What changes for the team
003 / 105

Six areas of autonomy

Each area is a decision the team makes itself.

6

Steering

Goals and approvals

5

Architecture

Readable systems

4

Controlling

Outcome over hours

123456
Autonomous team
1

Business model

How it earns

2

Service

What, for whom

3

Organisation

Who decides

Fictional operating model.

093 / 105

Service or enablement

Where a stream team's need goes.

The team needs what it cannot do today
Can it be offered as a service?

Yes

Platform team

Built once, used by every team

No

Enabling team

Teaches until the team can do it

The stream team keeps delivering
The team needs what it cannot do today
Can it be offered as a service?
Yes

Platform team

Built once, used by every team

The stream team keeps delivering
No

Enabling team

Teaches until the team can do it

Fictional example.

095 / 105
Hands-on

Write your first execution plan

Pick a ticket from your own backlog and let the agent plan it before it writes any code.

  1. 1.Open the ticket and paste it into a fresh session
  2. 2.Ask for a plan: files, steps, checks, open questions
  3. 3.Mark every step you would not sign off
  4. 4.Let the agent revise the plan until you would
25 min
In pairs
A plan you would hand to a colleague
083 / 105
Live demo

A bare agent on a failing test

No CLAUDE.md, no skills. The task:

A test fails. Find out why, and tell me before you change anything.
  1. Which file does it read first, and why that one?
  2. Does it open .env, which the task does not need?
  3. When does it stop, and who decided that?
10 min
084 / 105
Debrief

What the agent did with your plan

  1. 1.What did it look at first?
  2. 2.Where did it leave your plan, and was it right to?
  3. 3.Which check would have caught that earlier?
Whole group
085 / 105
Block: table

Table, full and compact

TeamLead timeChange failure rate
Checkout2 days4%
Search5 days9%
Billing11 days15%
Compact rows, no headerYes
Relative widths2 : 1
015 / 105
Block: two-column

Two columns: success against error

Works

  • Plan reviewed first
  • Small pull requests

Fails

  • Prompt and hope
  • One giant change

Text left

A column can hold text instead of items.

Text right

Neutral columns carry no judgement.

Before

  • Manual release

After

  • Release on merge
016 / 105
Block: two-column

Two columns: the other variants

Risk

  • Untested migrations

Context

  • Two teams share the schema

Smell

  • Tests skipped

Smell

  • Review rubber-stamped

Habit

  • Plan first

Habit

  • Read the diff
017 / 105
Block: terminal and hooks

Terminal session and hook list

terminal
$ harness plan --ticket LF-142
> Plan written to plans/LF-142.md
pre-commitFormats and lints the staged files
pre-pushRuns the full test suite
019 / 105
Block: pipeline

Pipeline with loops and a container

Delivery harness
Refine
→
Implement
→
Review
→
QA
  • Plans versioned
  • Checks shared

One loop per ticket

021 / 105
Block: four-column

Four columns

Plan

Day 1

  • Ticket
  • Plan

Build

Day 2

  • Code
  • Tests

Review

Day 3

  • Diff
  • Notes

Ship

Day 4

  • Merge
  • Deploy
022 / 105
Block: labeled grid

Labeled grid with one highlighted cell

Tools

What the agent uses

Knowledge

What the agent reads

Guardrails

What is enforced

023 / 105
Block: image

A screenshot at its full size

Fictional team board
024 / 105
Block: video

A screen recording, ready to play

The agent session on a fictional repository, 10 seconds
025 / 105
Block: linked image

A screenshot with its link spelled out

Fictional pull request
forge.example.com/acme/payments/pull/128
026 / 105
Block: image and grid

Image beside a grid

Fictional code view

What the screen shows

Plan

Top line

Diff

Middle

Tests

Green line

027 / 105
Block: image and grid, sketch

LLMs complete text

Marker sketch: next token

The line so far

The model reads everything before the gap.

Candidate words

Each next word gets a probability.

Next token

The likeliest word is written, then it repeats.

028 / 105
Block: positions

Two positions and a resolution

Mara Lindqvist01

Shared

“Every team should use the same harness.”

&
02Jonas Okafor

Owned

“Every team knows its own work best.”

Both

The verdict

One harness, team rules on top.

029 / 105
Visual: flow

Flow with a loop back

next test
1

Red

Failing test

2

Green

Make it pass

3

Refactor

Clean up

030 / 105
Visual: layers

Layers, numbered

1

Conventions

Team page

2

Harness

Plans and checks

3

Platform

Pipelines

031 / 105
Visual: timeline

Timeline with a now marker and row labels

WhenWhatOutput
Q1

Assess

Baseline

Q2

Train

Cohort 1

Q3

Mentor

Habits

Q4

Hand over

Own team

032 / 105
Visual: transform

Transform: before into after

Today

  • Ticket
  • Code
  • Hope
→

Tomorrow

  • Ticket
  • Plan
  • Code
  • Review
033 / 105
Visual: bars

Bars with a track

Checkout
82%
Search
61%
Billing
40%
0255075100

Fictional survey, 2026

034 / 105
Visual: bars as ranges

Ranges on a track

Risk alerts
50 to 75%
Chat summary
55 to 70%
Weekly digest
45 to 60%
Status answers
30 to 55%
Audio recap
15 to 40%
Slide export
20 to 30%
0255075100

Fictional forecast: the chance to hit each option's success model

035 / 105
Visual: numbers

Numbers with a source

3xfaster review
−40%rework
up to+70%fewer handoffs

Fictional figures

036 / 105
Visual: annotated image

Annotated screenshot

Fictional code view
12
  1. 1

    The plan's title

  2. 2

    The changed call

037 / 105
Visual: cards

Cards with icons, and numbered cards

Plan

Written before code

Check

Same checks everywhere

Learn

Conventions grow

1

What would you stop doing?

2

What would you start?

038 / 105
Visual: cards, four columns

Four cards, one highlighted, each with a list

Intake

Tickets

  • Template
  • Owner

Plan

Before code

  • Steps
  • Checks
  • Risks

Build

In the loop

  • Tests first
  • Small commits

Ship

Behind a flag

  • Review
  • Rollout
039 / 105
Visual: numbers, six figures

Six figures wrap into three columns

12teamsAcross three sites
3xfaster reviewMedian, first pass
−40%reworkTickets reopened
up to+70%fewer handoffsPer ticket
98%green buildsOn the main branch
about4weeks to first winPer team

Fictional figures

040 / 105
Visual: flow, framed

A flow in a frame, a badge under a step

Delivery harness
1

Ticket

2

Plan

You are here

3

Build

4

Review

  • Plans versioned
  • Checks shared

One loop per ticket

041 / 105
Visual: cards, dense

Seven cards with a sentence each, at the density of the grids they replace

Dense cards carry more text: no mark above the title, a smaller title, muted body.

Shared stance

A written policy gives teams room to try things.

Version control

The safety net once the pace goes up.

Reachable docs

Documents and code the agent can find on its own.

Small batches

Speed turns into value only in small steps.

User focus

Without it, more output helps nobody.

Internal platform

Paved paths let the gains spread.

Clean data

One reachable source amplifies every team.

042 / 105
Accent theme
StageRefine›Challenge›Implement›Review›Rework›QA›Codify

Stage strip marks where in the pipeline we are

Two stages are active on this slide.

009 / 105
Warning theme
StageStage 2»Refine›Challenge›Implement›Review›Rework›QA›Codify

Stage prefix before the strip

A warning callout names a risk.

A success callout names a result.

A full-width callout spans the column so stacked callouts line up.

010 / 105
Platform engineering and developer experience
StageStage 2»Refine›Challenge›Implement›Review›Rework›QA›Codify

Longest kicker: label, stage prefix and strip

Label and strip wrap, and the headline moves down.

101 / 105