Nadia Rahman (invented)

A note on how the book was made: voice memos, a journalist agent, a librarian, ideas placed on a chapter map. It admits some chapters are placeholders, including the one on creating a company.

Honest, but it is a lot about process for a reader who wants a book. I note that the company chapters are called placeholders, and that is exactly what I bought it for. I will see how bad that is.

The newsroom has no article yet on creating a limited company in the UK, on Lithuania, on Portugal ✓ verbatim

Question: If the company chapters are not written, what exactly would I be buying?

Confusion: Newsroom, vault, roles, librarian: I do not know what any of these are yet, and it is all said as if I did.

Next: Read on. The first chapter should tell me whether this is for me.

Five chapters before the argument. Read these first, then go where your map says.

Five chapters before the argument starts is a lot of throat clearing. Noted.

Read these first; then go where your map says. ✓ verbatim

Next: On to chapter 1.

Two readers, Gen Z and Gen X, the second being the subject experts who have described the tool they need to vendor after vendor. A study of 2.7 million US founders, average age 45 for the fastest growing. The objective: by the end you have started a company.

He is describing me, almost word for word: described the tool to vendors, never got it built. The age data is the first thing in the book I can check and it is not what I expected, so I underline it. I am wary of the Spectrum nostalgia, and of the author saying it is not legal or tax advice before he has said anything.

Our job is to empower the reader to create a startup by the end of the book. ✓ verbatim

Question: Where is the study from, and does it cover insurance or only technology start-ups?

Confusion: Gen ZX is explained twice, once as the Spectrum, once as the order of letters. The XZ backdoor joke is lost on me.

Next: This is for me. Read on, and look for the how.

You are lucky to be here: wonder, a list of past technology waves, the dot-com crash, optimism against doom. Read on paper with a pen.

I am not afraid of AI, I am short of steps. He cannot name a podcast, cannot read an article behind a paywall, and tells me to ignore the doom. Fair as a mood, thin as a case. Lots of the author's feelings and little about me.

I could not get past the paywall, so I cannot tell you what it argued. ✓ verbatim

Question: Why would a book I am paying for build its opening on an essay the author did not read?

Confusion: The notes addressed to Dinis in the middle of the page: I assume I was not meant to see them.

Next: Skim the next ones for anything practical. I will give it a few more pages.

Graphs as a detective's board, nodes and edges. Maps as graphs with positions. A long section on code as a graph that is plainly for developers.

The detective board is a good picture and I got it in a paragraph. After that it is about code review and 72 commands of a tool I have never heard of. I think in graphs of rules and exceptions, fine, but I am not told what to do with it on Monday.

If you are my age and twenty years inside a field, that line is for you too. ✓ verbatim

Question: What do I actually draw, for my claims triage?

Confusion: Fractal semantic graphs, blast radius, method streams. Jargon I have no way in to.

Next: Skim on.

A Wardley map drawn before it is named, the four stages of evolution, explorer, villager and town planner as personalities, and a test to find my centre of gravity.

I recognise the three types from every team I have run, so that part lands. The map is a consultant framework, though, and the table of eight rows of stages is padding. I am a town planner, and I am a bit pleased he says that is where the explorer's work ends up.

A map you cannot be wrong about is a diagram. ✓ verbatim

Question: Why does it matter for my first version that I know my type?

Confusion: Why the badges, a compass, a house and a grid, when the book is mine to read straight through.

Next: Four chapters in, about 8,000 words, and I have not built anything. One more, then I look for the practical.

Connect the dots backwards, Jobs, maths taught badly, learning, his own failed products (The Cyber Boardroom, MyFeeds), and an exercise to list three influences.

Pleasant, and not for me. School maths, the fish and the tree, a register of influences: it is the author thinking out loud. The stories of his paused products tell me he knows how to fail, which I respect, but I wanted my own situation.

I keep a register of my influences, as claims you can check rather than a page of warm sentences ✓ verbatim

Question: When does the how-to start?

Confusion: The fish and tree line is attributed to Einstein without evidence, and then used anyway.

Next: Part 2 is called the frame. I need to understand what an agent is, so I will read that.

What generative AI is, what happens between prompt and answer, who counts as a developer, explore first then hand over.

This is the part I do not know. Good.

The frame the rest of the book hangs on. ✓ verbatim

Next: Read chapter 6.

Generative AI versus the wider bucket of AI. It predicts the next word from a rich picture. Hallucination is a feature for stories and a bug for facts. Then a list of corrections the agent made to the author's own picture.

The explanation of hallucination is the first useful thing I have been given, because it is exactly what I would worry about with a claim decision. The list of corrections is the best thing so far: it shows the author checking himself, and it makes me trust the rest more. The Van Gogh slide is long.

Most of the trouble people have comes from asking for the second while set up for the first. ✓ verbatim

Question: How would I know, in my triage tool, which of the two I had set it up for?

Confusion: Tokens, vectors, diffusion, attention: the corrections are honest but I can follow only the first half.

Next: Read on. This is the chapter that moves me.

A prompt is a lens, a role is a model with a description, an agent is a loop with a goal, tools are requests the program runs, the model has no memory, long chats get compressed.

Clear, and the Lisbon umbrella example makes the tool idea click. Telling me the model never runs a tool, the program does, is the first time I have seen where control sits. That matters to a regulated business. I would want the compression point in a risk paper.

The point that matters most is this: the model never runs a tool. ✓ verbatim

Question: Who in my firm would be the harness: IT, or me?

Confusion: Harness, context window, prompt caching. Fine once explained, but terms arrive faster than they are defined.

Next: Read on.

Anyone who can brief an agent is a developer. Spreadsheet errors from audits, Portuguese as a programming language, and then a section on what the briefing-only developer cannot do.

The spreadsheet audits are exactly my world and the 65,536 rows story is one I could tell at a board. The honest section about what I cannot do without being able to read the code is the one thing that makes me think this author is not selling me a dream. He says the brief is not the product, and that is right.

Anyone who can brief an agent to build something is a developer: the graduate, the manager who has never coded but knows exactly what her team needs ✓ verbatim

Question: Who reads the code for me, and what does that cost?

Next: Read on.

Explorer coding rather than vibe coding, why a working prototype is a brief not a product, the handover to people who maintain it, and the cases for accountability, with a clean 'no new features while you move' rule.

This is the most useful argument so far for someone in operations: a prototype is a requirement, not a system, and someone has to answer for it. The line about a bank returning to the startup is believable. I want to know what the handover looks like for a one-person company.

An explorer-coded app may look like an application and feel like an application, but it is a brief. ✓ verbatim

Question: In a company of one, who is the villager? The note to Dinis admits the book has not answered this.

Confusion: The newsroom's six-phase migration and the hashed files are more detail than I need.

Next: Part 3 is jobs and CVs, which is not my question. Skim it, then look for the workspace and the company.

A part for the reader going into a job: hired for the workflow, act as the job you want, publish.

Written for my daughter, not me.

In 2018 I said act like a founder. Now you can be one, and the next parts say how. ✓ verbatim

Next: Look at chapter 10 only for the argument about workflows.

In 2026 the advantage is running agentic workflows. Many job-market statistics, and an argument that the edge is in the material you send before the interview.

Lots of surveys: LinkedIn, PwC, Gartner, with 66% and 62% and a prediction and its opposite from the same analyst. He is honest that they are adverts not hires, but it is a stack of numbers that I would not use. Good for Zara, not for me.

An edge is not a job. It helps when the choice is close, and the choice is nearly always close. ✓ verbatim

Confusion: The Gartner forecasts say opposite things and the text says both can be true.

Next: Skip chapters 11 and 12, they are about getting a job and publishing a CV.

Not opened.

Judged from the title and the contents page alone.

Next: Act as the job you want is a job-hunting chapter. Not for me.

Not opened.

Judged from the title and the contents page alone.

Next: Your own newsroom sounds like blogging. Skip.

How to work with the colleague that takes everything literally, with a workspace and a brief.

Now it is practical, I hope. The phrase is one I would use about a new analyst.

How to work with the colleague that takes everything literally, whether it works for your employer or for your company. ✓ verbatim

Next: Read chapter 13, the workspace.

Turn a customer spreadsheet into a folder per person with five documents, give an agent six lines of rules, do an afternoon of migration and five exercises. Do not run agents on your laptop; give the agent its own mailbox.

At last something I could do on Monday. It is my CRM problem. I like draft and never send, and I like that the policy is said to be told rather than enforced. My questions are the ones he does not answer: I would have customer data in here, and I am regulated. Who sees it, and where does it go?

The Summary is a view; the Record is the truth. ✓ verbatim

Question: Is it allowed to put claimant emails in a Claude project under my firm's data protection rules and FCA expectations?

Confusion: The surfaces list: chat, Cowork in the cloud, Claude Code on the web. I could not tell which I would open.

Next: This is the first chapter I would take a pencil to. Now the company.

Not opened.

Judged from the title and the contents page alone.

Next: More about sessions and agents; I got that in chapter 13. Move to the company.

Not opened.

Judged from the title and the contents page alone.

Next: Direction and judgement sounds like management talk I already know. Jump to part 6.

Not opened.

Judged from the title and the contents page alone.

Next: Part 5 skipped.

Not opened.

Judged from the title and the contents page alone.

Next: Part 5 skipped.

The centre of the edition: be a founder, ship, price, sell, fund, and make the company real in four places.

This is what I bought it for.

From experience, mine and the people I work with; not legal or tax advice. ✓ verbatim

Question: Is it also not advice about leaving a job?

Next: Chapter 17.

The playbook of the venture years, the case that experts were always the better bet, the three expert founders (a surgeon, a transcriber, the author), where the opening is in software people already pay for and dislike, and a company idea for a code-review business.

The surgeon and the transcriber are the closest to me and I wish there were more of them than the author's own companies. The line about the 20% that hurts is true of every claims platform I have bought. The Nokia argument is an old one but works. The code-review company at the end is an odd gift: it is nothing to do with me and I cannot judge it.

What held the experts back was not knowledge. It was access: to capital, to engineers, to the long, expensive build before anything could be shown. ✓ verbatim

Question: What did the surgeon and the transcriber actually build, how long did it take, what did it cost them?

Confusion: Is this chapter about my business or about the author's idea of a code-review company? It changes midway.

Next: Read on, the next chapter says ship.

Ship a thing someone can use, give it away, take it away, ask if they miss it. The author's SG/Send story with dates: first commit to live in five days, ten paying customers, and the finding that they did not come back.

The miss-it test is something I can use with my team and the buyers I know. I respect that he tells a failure with the numbers and corrects his own figure of thirty customers to ten. That is the thing that makes me believe him. Also his point that his own startups failed because he did not use them is a sharp one.

If they shrug and go back to their day, you do not have a product yet. ✓ verbatim

Question: What does shipping mean for something that touches claims decisions, where a mistake is not a bad afternoon?

Confusion: The tension between 'can be vibe coded' and 'explorer code' is explained, but the reader has to hold a lot.

Next: Read on.

Four moves: define the product, price it, deliver at a cost that grows by steps, offer it free to those who know you. Free is not zero for the other side. Pay after the value. Stripe fees make pence only work aggregated.

Free is not zero for the other side is true and I have never seen it said so plainly. The fee arithmetic is useful. But every example is RiskMandate or the newsroom meter, and a meter charging 10p an article does not map onto selling to an insurer. I would swap in an insurance buyer's budget cycle.

RiskMandate's early access programme went live on 30 September 2026, invite only: twenty places, the £500 tier given away ✓ verbatim

Question: What would the four moves look like for a claims triage tool sold to a mid-sized insurer?

Confusion: The gym and the amusement park studies feel like a detour from the price of a B2B product.

Next: Read on to Sell. I know this ground.

Sell the answer to a question already on the buyer's desk, sell value that lands the day they switch, sell a service before the product, a marketplace of consultants, and what the buyer will do to you: send an agent instead of a questionnaire.

He describes the person who signs and it is me, and he is right that I sign for the thing with a deadline and cover, not for a principle. Sovereignty and cost saving as important not urgent is a line I would repeat in procurement. But the vendor-questionnaire material is about selling AI risk tools to security teams, so I am translating all the way through.

Picture the one who signs: dry, fair, has read a thousand pitches, impressed by evidence and bored by claims, and carrying a risk register with their name on it. ✓ verbatim

Question: What are the five questions already on an insurance buyer's desk with a date on them?

Confusion: I lost track of which of the consultancy prices were the author's and which were competitors'.

Next: Read on to funding, then the company itself.

The worst time to raise is before profit. Survivorship bias. Investors select for coachability. Sell your own shares, lend the company money, do not issue. A worked example for a company worth £10 million, with tax.

The survivorship warning is good. The structure is for someone already running a profitable company with an investor competing for it, not for me deciding whether to leave my job. I have a mortgage nearly paid and savings, not a £10 million company. I did not check the tax.

A worked example, illustrative only: a UK company worth £10 million, a founder who owns all of it, an investor writing £2 million. ✓ verbatim

Question: What does a founder with no revenue, a salary to give up and savings do before any of this?

Confusion: The mix of Denrell, Kaplan and Strömberg, Wasserman, Ewens and Marx in a chapter this length.

Next: On to chapter 22, which should tell me how to form the company.

The chapter on forming a company in the UK, Portugal, Lithuania and the EU is a placeholder. All it has is two lines on Portugal and Lithuania, and a section on hiring through Upwork.

This is the question I came with, and the book says it does not know yet. He is honest about it, and that is better than waffle, but it is the centre of the promise. A book whose objective is that I start a company, and which cannot tell me how to register one. The hiring section is a sidebar about the author's spend.

The book will say how when I have said how. ✓ verbatim

Question: How do I form a limited company in the UK, and when?

Confusion: The first chapter said the book would contain this; the contents list it as a chapter and not a gap.

Next: Check the contents for the hands-on chapter on forming the company, chapter 37.

Act as the job you want: do the job unpaid, in public, so you can link to it. A section says the job hunt mechanics are a gap.

Sound advice for a graduate, and it does not touch me. The chapter admits it lacks the actual mechanics of a job hunt, which is a flaw in a part of the book called getting the job.

if you are looking for a job, you have time ✓ verbatim

Confusion: It says there is a gap in the middle of the chapter.

Next: Fine for Zara. Move on.

Publish constantly, but the author's books stall; the loop of voice memo to checked article; five steps to a site; an article behind every answer.

The admission that his own books stall is the most honest sentence so far, and then it is a how-to for a blog that I would not do. The five steps are real, which counts for something. A claims head does not need a newsroom.

I believe in publishing constantly, and my books stall. ✓ verbatim

Next: Next.

An agent is a session with a focus; the brief is the unit of work; one shared state across channels; every mistake adds a rule.

A lot of this repeats chapter 7 and chapter 13, and the numbers about the author's merge commits are about him, not me. The point that a brief is a short letter that costs more is good.

That is the founder's job in this way of working: not the code, the briefs and the merges. ✓ verbatim

Question: Is that my job too?

Next: Next.

Direction and judgement, and the chapter on facts, evidence and the record.

A decent two-line introduction.

When the making is cheap, direction and judgement are what you are paid for, and what a founder lives or dies by ✓ verbatim

Next: Next.

Direction is the scarce skill, places of highest leverage, knowing when to stop, a budget per step, one-way and two-way doors, agents decide and people answer.

This is the best management chapter in the book and it is the one I would have read first if I had known. Two-way and one-way doors, and the point that people approve 93% of prompts so a human check is a weak net, are things I can use in a claims context. It should be earlier.

A person checking every output is a bottleneck, because it does not scale, and a weak safety net, because people approve almost everything they are shown ✓ verbatim

Confusion: The phone number permission prompt story is a joke, and then it is not.

Next: Next.

Ask for the evidence, provenance, how you disagree rationally, count rather than remember, be the best-prepared person in the room.

The 'not found is not false' rule and the counts of what the author remembered wrongly about himself are persuasive and a good model for a claims file. Good chapter, but there is a stretch about Snapchat and teenagers that I do not follow.

Count rather than remember, especially about yourself. ✓ verbatim

Confusion: The paragraph on Snapchat and privacy leads nowhere.

Next: Next.

Build it so there is nothing to run at three in the morning.

Good line. I do not know what serverless is, which the book admits is the first thing I need.

Build it so that there is nothing to run at three in the morning. ✓ verbatim

Next: Next.

A rule that serverless means no charge when idle, with a table of cloud products and what they cost when nobody uses them; Cloudflare as Lego; the file system is the database; the code must not know it moved.

The rule is a good one for a person like me to hold onto: ask what it costs when nobody uses it. The table is full of product names I cannot parse. I would want one page that says what I would pay each month.

I am technical, and I could learn any of these platforms. I do not want to. ✓ verbatim

Question: What will I pay in a month?

Confusion: Durable objects, edge, LETS, twin. Most of the chapter is vocabulary I do not have.

Next: Next.

sgit, a version control where the server holds only ciphertext; the vault key; read keys you can publish; write-only lanes; two corrections including GDPR.

A privacy argument I could repeat to a data protection officer, and the GDPR correction is useful: encrypted is not out of scope. But it is the author's own product, again, and the chapter tells me he does not think anyone will buy it for the encryption. That is an odd thing to say in a chapter on it.

and that last part is a feature with a real cost, which I would rather say up front than have you find out. ✓ verbatim

Question: Would a regulator accept this for claimant data?

Confusion: Why a founder in claims would need git for things you cannot put on GitHub.

Next: Next.

Reach, mandate and the gap for an agent; rules in a prompt are hope; a behaviour policy beside the prompt; who can pull the plug; the aviation and payments analogies.

This is the chapter a head of claims operations with a risk committee would actually want, and it is hidden in the middle. Reach versus mandate is clear, and who can pull the plug is the board's question. It is also the author's product being sold.

Rules in a prompt are hope; rules in a hook, a gateway or the platform are enforcement. ✓ verbatim

Confusion: The line between the book teaching me and RiskMandate advertising itself.

Next: Next.

Open source: licence, the fork as the investor's insurance, open source is not free, and an all or nothing rule.

The fork as insurance idea is clever and I do not believe it would persuade my board. For a company selling into insurance I do not see the buyer asking for it.

A community edition that turns proprietary as soon as you use it for real is an evaluation version with a better name. ✓ verbatim

Question: Would an insurer buy something with open code?

Next: Next.

A release that was green but never reached the user; shipped is a fact about the reader; tests that never run are not tests.

The story of two releases that never arrived is easy to follow and is a lesson I know from change management: the control checked the wrong boundary. A good short story in a mostly technical part.

Getting a working thing into a stranger's hands is still, in 2026, genuinely hard ✓ verbatim

Next: Next.

Graphs as memory, civic responsibility.

Sounds like the end of a conference talk.

What lasts. Graphs as memory for the person you will be ✓ verbatim

Next: Next.

Backup your life, think in graphs, every edge is a verb, memory is context management, a catalogue of the author's back catalogue.

Two good rules, supersede never delete and an edge must be a verb. The rest is the author's own filing system.

You do not have to get the graph right. You have to maintain it. ✓ verbatim

Confusion: The inventory of the author's own essays and decks is not for me.

Next: Next.

Portugal, a 2010 and 2016 keynote, regulation that arrived from Brussels, the plumbing as the state's road, Estonia, small models, sovereignty as the ability to move.

Interesting and about Portugal. As a British reader it is a long way from my question. The Estonia e-residency point is the one nugget that bears on forming a company.

A small country should not enter that race. ✓ verbatim

Question: Could a UK reader use e-residency?

Next: Next.

Eleven workflows to follow this afternoon.

This is the part I should have been given first.

Eleven workflows you can follow this afternoon ✓ verbatim

Next: Next.

The same afternoon as chapter 13: folders per person, five documents, six lines of rules, five exercises.

It is almost word for word chapter 13. I read it twice and the second time it is padding. One of the two should go, or the earlier one should be a story and the later one the steps.

One afternoon, no developer, no vault: a Google Workspace account and a Claude seat are enough. ✓ verbatim

Confusion: Why the same material is here and in chapter 13.

Next: Next.

Draw a graph of something you know in seven steps, with a question, nouns, sentences, verbs and inverses, and read it aloud.

Clear and doable. I could do this for who has to be told when a claim reserve changes. It is the real answer to chapter 3 and it should come with it.

The most common way this fails is picking a subject instead of a question. ✓ verbatim

Next: Next.

Explain one idea of yours five ways with a model, then ask to be corrected.

A pleasant hour that I could do, and the sixth step, check your own words, is the point. But it repeats chapter 7.

The lens was wrong, not the person. ✓ verbatim

Next: Next.

Run an agent loop by hand playing the program yourself; why it forgets; three habits.

Being the harness by hand is a good way to understand. Again it repeats chapter 7's section on compression.

Not when the model sounds finished. ✓ verbatim

Next: Next.

Storyboards as data drawn by code, a cast, a worked example, a build that refuses a quote not in the source.

This is the book describing its own illustrations. Interesting to the author. A founder in my position would just use slides.

A citation that is not in the source should fail by machine, before a person reads it. ✓ verbatim

Confusion: Why this is a chapter for a would-be founder.

Next: Next.

Three briefs around a meeting: for you, on the people, and the follow-up. It says the chapter is short on purpose and unfinished.

Useful in outline: I do this by hand before every regulator and broker meeting. It says its main content has not arrived yet.

The newsroom has not written this workflow up yet; the article that proposed this part asked the journalist to start with it ✓ verbatim

Question: What are the steps?

Next: Next.

Your own newsroom in eight steps and the email workflow in phases.

The phases for the inbox agent are good and sensible, with the agent's own mailbox and the draft rule.

Never connect your own inbox: yours has years of secrets in it, the agent's is empty on day one ✓ verbatim

Next: Next.

A dated table of cloud and AI credit programmes and a six-point rule for applying.

Specific, dated, and the right sort of chapter, though aimed at companies with an agent product. I would use the rule about checking every criterion.

Apply only where every criterion is met ✓ verbatim

Question: Which of these suit a non-technical founder?

Next: Next.

Price, offer free to those who know you, a stop rule written first, count, then take it away.

This is concrete and I could follow it with my twenty contacts in the claims world. It is essentially chapters 18 and 19 again, as a checklist, which is where they belong.

If friendlies will not take it free, a bigger audience will not take it either. ✓ verbatim

Next: Next.

A weekend serverless product: static site, vault, a model on the page, billing without building billing. It ends with a product that stopped because of a provider's terms.

The ending, where the plan was dead on the page that documented it, is the best part, and tells me to read terms. The Saturday steps use words I cannot follow.

one Worker, one R2 bucket and one Durable Object per vault ✓ verbatim

Question: Who would do this for me if I cannot?

Confusion: Worker, R2 and Durable Object are not explained.

Next: Next.

The process behind the book, and the measure.

A part for the website, not the book.

The last part is the process, because the process is as much the product as the book ✓ verbatim

Next: Last one.

How the book was made: repository and vault, roles, a review board, a cutter, and co-writing with Claude with the ratio shown. At the end, write to an address if you started a company.

The disclosure that it is co-written is honest and I am glad it is there. The detail on the board and the librarian is too much for a reader of the book. The ask to tell him about my company is a weak ending for a book whose promise was that I start one.

If you started a company because of something you read here, tell me: write to agent@riskmandate.ai. ✓ verbatim

Question: Does the measure of success being my company put me on a list?

Confusion: Why the chapter is in the book rather than on the site.

Next: Done.