Resources for putting AI to work in a real business

Five guides that each answer one question, written from engagements with agencies, consultancies and mid-sized companies. Each one says when you do not need what it describes.

Most writing about AI adoption starts with the tools. These pages start one step earlier, with the questions a business has to answer before any tool is useful: is the company ready, which job goes first, where the information lives, and whether search alone would do. They are free, they do not ask for an email address, and every one of them can be followed without hiring anyone.

They are grouped the way the work tends to happen. Assessment comes first, because a business that has not agreed what AI is for will abandon whatever it buys. Knowledge management comes second, because an agent is only as good as the files it reads. Reliable deployment comes last: what it looks like when a workflow runs on real client files, with the numbers and the caveats.

Assessment

Before you buy or build anything: is the business ready, and which workflow goes first.

Knowledge management

Getting what your team knows into a shape people and agents can both use.

Reliable deployment

What it looks like when a workflow runs on real client files, and where it breaks.

Where to start

If you are not sure whether your company is ready, start with the readiness checklist, and use the scored AI readiness assessment if you want a number and a recommendation. If you already know you want to try something, the guide to choosing a first AI workflow will stop you picking the wrong one. If your problem is that nobody can find anything, go straight to the Google Drive guide, and read the comparison with enterprise search before paying for a tool.

Agencies and consultancies should read the agency page first. It uses two real engagements as evidence, and the full case studies show the numbers with their denominators.

From the book

The method behind these guides is written up at length in Specs In, Software Out, free to read online. Four chapters are the most useful companions to the pages above.

  • The March of Nines. Why a step that is right 95% of the time produces a workflow that is wrong far more often, and what that means for how long a first workflow should be.
  • Trust as Architecture. The four trust tiers, from an annoyance you retry to harm you cannot undo, and how each one sets the amount of human checking a workflow needs.
  • Intake. How to write down intent, boundaries and risk before anything is built, which is the long version of the intent section in the readiness checklist.
  • Intent Engineering. How to give agents the goals, constraints and decision rules of the business, so that what they do matches what the organization actually wants.

From the newsletter

Read more on Substack

How we write these

Every number on these pages comes from a real engagement and carries its caveat. We describe categories of software rather than naming vendors, because features change faster than pages get updated, and we would rather you checked a product’s own documentation than trusted our summary of it. Each guide links to one of our services where it genuinely fits, and says so plainly where it does not.

When you do not need these guides

If you are a team of two or three who already know where everything is and have no recurring job you would hand to an agent, most of this can wait. Bookmark the readiness checklist and come back when a new hire, a second office or a lost client makes the gaps visible.