Agentic AI in practice: how Sage is reshaping accounting with MTD

Signal note — full article published on Sage: Everyday agentic AI: How Sage is reshaping the future of accounting with MTD

This article is significant not as a product announcement but as a statement of the operating model shift. The MTD for Income Tax Agent makes the capacity gap argument concrete: agentic AI doesn’t help accountants work faster, it restructures what accountants are for.

The framing from Chris Downing, Director for Accountants and Bookkeepers at Sage:

“Capacity creation isn’t a side effect of automation. It’s the strategy.”

That’s the AI Maturity Curve in a sentence. Not productivity gains — structural reallocation of what the human does.

The collision points — precisely dated

The article lays out the compliance calendar that creates the pressure:

7 August 2026 — first quarterly update for £50,000+ sole traders and landlords
April 2027 — £30,000 threshold; payrolling of Benefits in Kind replaces P11Ds; mandatory
April 2028 — £20,000 threshold
Ongoing — Companies House digital filing; VAT cycles; overlapping quarterly workloads

This isn’t a single transition. It’s a compounding series of changes, each one narrowing the window for firms still operating on episodic models.

Emma Rawson, ATT Director of Public Policy: “Twelve months ago, we were convincing people [MTD] was happening. Now the conversation has shifted to readiness.”

What the MTD agent actually does

The MTD for Income Tax Agent operates across four areas:

Preparation and authorisation — anticipates what needs doing, prompts action on HMRC Agent Services Account linking and authority verification. Authorisation with HMRC remains manual.

Guardrails by design — accountants decide how far automation runs. High-confidence tasks are automated; anything ambiguous is routed for review. This is the glass box model applied to workflow — not autonomous action, but supervised execution.

Quarterly updates with context — the agent reviews incoming information, highlights anomalies, requests missing items, and submits once the accountant approves.

Guided client experience — for clients who won’t adopt software, the agent provides simple approval and upload flows. The default motion remains bringing clients into software for full visibility.

Georgina Timothy, Director of Product Management, Sage for Accountants: “We’ll automate the end-to-end workflow, but you stay in control. The agent takes care of the heavy lifting so you can focus on your clients.”

What changes in tax returns

This is the structural shift that most MTD coverage misses.

In the new model, HMRC generates estimated tax calculations based on quarterly data. Accountants don’t generate the numbers — they validate and interpret them.

Downing: “The software providers aren’t doing the tax calculation. HMRC is. Your job becomes validating and reconciling that information, making sure what HMRC holds matches what your clients have.”

The tax return is becoming a continuous data exchange rather than an annual event. The BSAS (business source adjustable summary) ties quarterly updates to final tax adjustments — the mechanism that keeps client records and HMRC’s view in sync.

Less time keying figures. More time shaping what they mean.

This is the accountability concentration dynamic the glass box article describes — applied specifically to tax. The moments of human judgment become fewer, more consequential, and more visible.

The five-step playbook — the pricing signal

The article’s recommended sequence is analytically useful:

Segment and sequence clients by working style and income threshold
Digitise records early — don’t wait for MTD to force it
Automate the quarterly cycle — identify what can safely be handed over
Price the new service model — package quarterly reviews into modern service tiers, update letters of engagement now
Reinvest the capacity created — ring-fence hours for advisory conversations

Step four matters most. The same sequencing the MTD Pricing Framework proposes: commercial model change before operational change, because without repriced engagements, efficiency gains have no mechanism to become margin.

Rebecca Benneyworth: “Block out your quarter-end weeks now. This isn’t four mad months and eight quiet ones — it’s a continuous rhythm.”

Authentic Intelligence — the product positioning

The article closes on the “Authentic Intelligence” framing — people and agents, each doing what they’re best at.

This is Trusted AI applied to workflow: not AI replacing accountants, but AI restructuring what accountants do so that the human contribution is judgment, relationships, and strategy rather than reconciliation, submissions, and chasing.

Downing’s version: “The question isn’t how we cope with more deadlines. It’s what we do with our capacity.”

That framing resolves the anxiety that surrounds agentic AI in professional services. The agent doesn’t displace the accountant — it creates the conditions under which the accountant’s judgment is the valuable thing, rather than the time spent producing outputs that the agent can handle.

Connection to the Hidden Hours research

The article names the hidden work explicitly — “the chasing, corrections, and compliance housekeeping that consume hours every week” — echoing the Hidden Hours research vocabulary.

The 81% performing work outside agreed scope. The 70% whose fees don’t reflect the full support they provide. MTD’s quarterly rhythm creates more touchpoints, more chasing, more coordination. Without agentic workflow to absorb the operational layer, the hidden hours expand further.

The agent makes the argument for repricing: if the compliance layer is automated, the remaining human work is advisory — and advisory work is priced differently from compliance work.

That’s the commercial opportunity embedded in the operational shift.