Hi

When my book 'Practice! How Meera built  an accounting practice that gave her more money, financial and time freedom' came out in March 2025, AI was the shiny new tool everyone was experimenting with.

“Look,” we all said, “it can write an email.” Useful? Yes. Transformational? Not quite.

Fast forward only nine months to now and the conversation has changed. The question is no longer “Can AI generate content?” It’s “Can AI run parts of my practice?”

That’s why I’m writing a Second Edition sooner than planned. Not because the first edition was wrong. It wasn’t. It was right for its moment. But the moment moved.

And if you’re building (or rebuilding) a practice today, you can’t afford to follow a 2025 playbook in a 2026 profession.

The pace of change isn’t slowing down

Accountants are used to change. We’ve lived through cloud accounting, Making Tax Digital, bank feeds, automation, client portals, and a steady stream of “new HMRC processes” that arrive like uninvited houseguests.

But AI is different.

Cloud changed where the work happens. Automation changed how quickly the work happens. Agentic AI is changing who (or what) does the work — and what “doing the work” even means. If that sounds dramatic, good. 

The big shift: from generative AI to agentic AI

In 2025, most firms used AI like a clever assistant: draft this email, summarise this meeting, rewrite this paragraph, suggest a marketing post, produce a checklist.

That’s generative AI. It produces words, structure, ideas. It’s a productivity booster.

Now we’re moving into agentic AI — systems that can pursue an outcome by taking steps, using tools, and moving workflows forward, with approvals, if you set them up properly.

It’s the difference between: “Help me write this client email” and “Chase the missing records, categorise what arrives, flag exceptions, draft the client questions, and update the workflow.”

Why a Second Edition (instead of leaving it alone)

Here’s the uncomfortable truth. If I leave the book untouched, some readers will use it as a blueprint for how to build a modern practice — and they’ll build a practice that is “digitally tidy” but operationally outdated.

Not because they’re incompetent. Because they followed the instructions. A responsible book has to match the world it’s written for. So the Second Edition is not a vanity project. It’s a reality update.

The biggest changes you’ll see in the Second Edition

I’m not rewriting the whole book. The fundamentals still matter:

• good quality clients
• clear scope of work
• fixed pricing
• clean processes
• a sensible tech stack
• professional judgement

But the AI layer now needs to be treated as part of the operating model, not a side note. Here are the biggest changes.

1) The book becomes “AI-first, human-led”

This is the phrase that matters. AI-first doesn’t mean reckless. It doesn’t mean replacing humans. It means designing the practice so that routine work is handled by systems, and the human work is where it should be:

• interpretation
• judgement
• explanation
• challenge
• advisory
• accountability

If your practice still relies on manual effort for routine work, you’re not “traditional”. You’re exposed.

2) David’s “future of the profession” talk gets sharper (and more honest)

In the first edition, parts of the AI discussion can read a little too clean: the implied promise of “instant” and “error-free”. That’s not how professional work happens. The update makes the real point that the technology will do a lot but it won’t be perfect. The winners won’t be the firms who pretend it’s perfect they’ll be the firms who build exceptions, approvals, and evidence into their workflows

Autonomy without governance is just a faster way to make mistakes.

3) Principle 5 upgrades from “Digital Practice” to “AI-First Practice”

The Seven Principles stay. The structure stays. The message deepens. An AI-first practice is built on three non-negotiables:

• Design for exceptions: systems handle routine, humans handle judgement
• Single source of truth: messy data produces messy outputs — quickly
• Controls beat cleverness: permissions, approvals, audit trails, review routines

If you want to use agentic AI, you don’t start with prompts. You start with design.

4) The compliance chapter gains an AI governance section

This is where many firms will come unstuck. The Second Edition adds practical guidance on:

• what data can and cannot be used with AI tools
• how to treat AI as a supplier
• what “review” means in an AI-assisted workflow
• how to build an audit trail habit so you can defend your work

If you can’t explain how an output was produced, you can’t defend it. And if you can’t defend it, you shouldn’t be sending it.

5) The tech stack chapter adds “the AI layer”

In the first edition, the focus is rightly on choosing a platform, simplifying your stack, and integrating properly. That still matters. Even more now. But the Second Edition adds a simple, practical distinction:

• Copilots help you create and summarise
• Agents help you execute multi-step tasks
• Orchestration is what stops it all becoming chaos (rules, approvals, logs, exception handling)

You don’t need every tool. You need one workflow that works. Then another.

6) The AI chapter is rebuilt around operating model, not prompting

Prompting still has value — as a thinking tool. But it’s no longer the main event.

The revised chapter is built around where agentic AI fits in a practice, what it should and should not do, what guardrails are non-negotiable, how it affects service delivery and pricing and why “outcomes over hours” becomes even more important.

7) New templates: policies, checklists, and incident planning

Because good intentions don’t scale the Second Edition adds five print-ready one-pagers:

• AI Acceptable Use Policy
• AI Supplier Due Diligence Checklist
• Agent Workflow Design Sheet
• AI Review Checklist (deliverables)
• AI Incident Plan

Not exciting. Completely necessary.

What this means for you (practically)

If you’re building or scaling a practice in 2026, here’s what I want you to take away:

• Stop thinking of AI as a feature. Start thinking of it as part of your operating system.
• Don’t chase novelty. Design one governed workflow that removes real friction.
• Raise your standards, not your speed. Faster work without controls is not progress.
• Price the outcome. The more efficient you become, the less sense hourly pricing makes.
• Keep your judgement centre stage. That’s the part clients can’t buy from a machine.

A final thought

This Second Edition is being written “so soon” because the profession is changing in real time. And whether you love AI, hate it, or would rather ignore it, it’s still going to change the expectations your clients bring to the relationship.

The good news?

This isn’t about becoming a tech company. It’s about becoming a better practice - clearer, more controlled, more valuable — and far less dependent on you doing everything yourself.

If that sounds like the practice you want, you’ll like what’s coming in the next edition of Practice!

Noel Guilford