This Week in Brief

Last week’s Mills Review reaction pieces have given way to harder data and practical guidance. New survey research shows advisers have effectively drawn their own boundary around agentic AI — comfortable handing over admin, far more cautious the moment client money is involved — while separate voices from AdvisoryAI, Dynamic Planner and Parmenion published concrete rules for building AI that’s actually safe to put in front of clients.

Key Developments

Advisers Will Use Agentic AI — Just Not on Client Money

Update: New research from GBST and the lang cat, surveying 178 UK advisers, puts numbers on the boundary line first flagged in last week’s Mills Review reaction pieces. Eighty per cent are comfortable with AI collating data for annual reviews and suitability packs, but comfort drops sharply for anything touching client money directly — only 43% are happy with AI on pension transfers, 53% on CIP switching and rebalancing. Nearly a third of advisers still can’t describe what “agentic AI” actually does, suggesting the discomfort may be partly a familiarity problem rather than a considered risk judgement. (FT Adviser, 20 August 2026)

“A Document Library You Cannot Query Is Just a Warehouse”

AdvisoryAI’s chief executive Alan Gurung made the case for AI that lets advisers ask questions of a firm’s own documents, rather than just generate new ones — for example, instantly surfacing which clients have fact-finds over a year old, turning a compliance sampling exercise into a single query. He was clear about the guardrails: the tool should read files, not silently rewrite them, and any change to a client record should be a deliberate human decision. For firms sitting on years of suitability reports and meeting notes, this is a useful reframe of what “AI for admin” can mean beyond drafting. (FT Adviser, 19 August 2026)

Dynamic Planner Sets Out What “Advice-Grade” AI Should Look Like

Louis Williams, head of psychology and behavioural insights at Dynamic Planner, argued that AI agents built for advice need to clear a higher bar than consumer chatbots: transparency about when AI was involved, framing that’s checked against a client’s actual goals, and a genuine human review rather than rubber-stamping. His practical worry is that an AI agent built without behavioural-science input could easily reinforce the same biases — like over-weighting recent fund performance — that advisers exist to manage. Worth reading for firms evaluating any agentic AI tool, as a checklist of questions to ask vendors. (Professional Adviser, 17 August 2026)

A Five-Step Guide to Building a Compliant Website Chatbot

Parmenion’s Chris Falconer set out a simple framework for firms experimenting with a client-facing chatbot: be explicit it’s an assistant, not an adviser; keep its scope narrow (product info, booking, simple explainers); set clear rules for when a human takes over; and add plain disclaimers that nothing it says is financial advice. It’s a useful starting checklist for smaller firms tempted to bolt a chatbot onto their website without first working out where the guardrails need to sit. (Professional Adviser, 19 August 2026)

From the Trade Press

  • Finance workers want a seat at the table on AI rollout — A survey of Aegis Union members found 85% had already seen AI or new technology introduced into their area of work, but only 38.5% said their employer had consulted staff first. The union has launched a charter campaign asking employers to give workers a meaningful say in how AI changes their jobs. FT Adviser, 21 August 2026
  • Is the AI pricing bubble about to burst? — An opinion piece questions how long AI providers can keep selling access below cost to build market share, noting most current business use looks like exploration rather than genuine dependency. Relevant for any firm banking on today’s AI subscription prices staying low. FT Adviser, 20 August 2026

What to Watch

Keep an eye on AI pricing. Most AI tools advisers rely on today are sold below cost while providers chase market share — if that changes, subscription costs for AI-powered adviser software could rise faster than firms are budgeting for. Separately, watch for more firms publishing their own “advice-grade AI” criteria, as vendors come under pressure to show their tools meet a higher bar than general-purpose chatbots.

Sources: FT Adviser, Professional Adviser, web search. Compiled 21 August 2026.