Never send a human to do a machine's job

29 Apr 2025

Let’s prioritise the backlog. Ready when you are Mr. Anderson…

tl:dr

1. Explicit: AI is already moving from acceleration to autonomy
2. Implicit: It’s going to change both Product Delivery and the Financial Services business model
3. Tacit: Adaptation will be unavoidable
4. Beyond the event horizon… there’s always opportunity

Intro and Warning:

Let me get the remaining AI generated meme references out of the way first. I promise then it’s just very clever - walls of text all the way down.

I recently completed a Product x AI Playbook detailing the applications, workflows and implementation of AI as both a way to expedite delivery across the Development Lifecycle (Discover, Validate, Build, Launch, Evaluate, Iterate) and for building AI features and products. It was changing almost daily with news of new ways to embed, extend or orchestrate. It felt a lot like the picture on the left above and by the end of it I felt like the picture on the right.

1. Mirror - Signal - Manoeuver

The first wave of Generative AI (GenAI) in Product was about acceleration. It enables teams to compress weeks of effort into days, replacing grind with automated leverage.

Every product manager should recognize these shifts:

• Before AI:
  • Monday: Gather scattered inputs (2 hrs)
  • Tuesday: Blank page to first draft (3 hrs)
  • Wednesday: Stakeholder feedback (2 hrs)
  • Thursday: Endless revisions (1+ hr)
  • Friday: “Actually, can we add…” (∞)
• With AI:
  • Dump context into a prompt (5 min)
  • Review generated draft (20 min)
  • Customise (20 min)
  • Ship

Research followed the same pattern. What once took for example +19 hours across 2 weeks (recording, transcription, clustering, writing insights) can now be reduced to about 10 hours in a few days. Slide creation collapsed from 7–8 hours of blank-slide paralysis to 2–3 hours focused on design and delivery.

The playbook codified these gains: especially regarding GenAI as a workflow enhancer. It was pragmatic and useful, but it also assumed a constant: humans would remain the ultimate decision-makers.

That assumption no longer holds. As much as I hate to say it, Agentic AI represents the next phase. These systems don’t just generate; they decide and act. Leaving aside for a moment the impending “Product Managerbot”, it’s pretty obvious that the move from co-pilot into participant extends to markets themselves.

• A TravelBot won’t just draft an itinerary, it books it.
• A SaveBot won’t just recommend a balance sweep, it executes it.
• A finance agent won’t just suggest the best card, it switches transactions in real time.

Whether by Agent Smith or yours truly, products will increasingly have to be designed for an environment where AI agents can transact, optimise and arbitrate on behalf of customers.

2. Not so fuzzy Margins

Financial Services profitability has long rested on what could be politely called the subsidy of inefficiency:

• Low-yield deposits left untouched.
• Credit-card rewards unredeemed.
• Fees collected because switching was too hard.

This model works because customers don’t optimise every dollar, every day.

Agentic AI changes that. Optimisation becomes continuous and automatic. Margins once protected by opacity are exposed.

Where McKinsey described this as the “collapse of inertia,” it might as well be called the end of margin opacity. The subsidy disappears because agents act relentlessly in the customer’s interest.

Examples are already here:

• Deposits: agents sweeping idle cash to higher-yield accounts and back again just in time for bill payments.
• Cards: agents auto-selecting the best card for each transaction, rotating balances before promotional rates expire.
• Payments: account-to-account (A2A) transfers bypassing interchange altogether.

For banks, this means pressure on low-cost funding and liquidity assumptions. For card issuers, it means erosion of interchange and reward arbitrage. For merchants, it means a mini-auction at every checkout where agents choose the least-cost rail.

Products cannot rely on hidden spreads or consumer inertia for profitability. In an agent-mediated market, value must be earned visibly, transaction by transaction.

3. Is that for me?…

If agents are becoming active participants, then product design must adapt. Loyalty will still matter, but it will increasingly be mediated by algorithms optimising for outcomes.

That requires FS products to become AI-facing as well as human-facing:

• Discoverability: machine-readable APIs, transparent pricing metadata, standardised terms.
• Interpretability: guardrails, audit trails, and explainability hooks so agents (and regulators) can trust outputs.
• Optimisability: structured value metrics that agents can use to rank choices.

In effect, this is SEO for agents. Just as websites were redesigned to be legible to search engines, FS products must now be designed to be legible to AI agents.

The product manager’s task changes. It is no longer just about human UX, although that still matters for brand affinity and trust. It is about building dual-facing products: experiences compelling to humans but also optimisable by algorithms.

Strategic questions to ask:

1. If inefficiency disappeared tomorrow, would our product still win?
2. Are our products visible and selectable by AI agents?
3. Which control points in the agentic economy (data, consent, trust, integration) can we credibly own?

Product strategy must extend to mapping how AI systems will discover, parse, and act on product information.

We’re not in Kansas anymore

When AI lowers the cost of creating code and features to nearly zero, speed is no longer the scarce resource. The scarce resource becomes judgement.

This demands a new craft for Product Managers: what could be termed focused restraint.

• Experiment widely using lots more data but cut ruthlessly.
• Invest in features that matter both to human trust and to agent logic.
• Balance automation with areas where emotional resonance or human reassurance still creates stickiness.

Differentiation in this world won’t come from owning the most advanced model as those are commoditising fast. It will come from:

• Proprietary data assets that agents cannot replicate.

• Integration depth into payment, compliance, and identity rails.

• Trust and liability wrappers that reduce risk for both customers and agents.

• Human-layer experiences that create emotional preference beyond raw optimisation.

The Product Manager’s timeless question remains: What do I build, and why will it win?. But the answer must now be valid for both a human customer and the AI agent representing them.

The future craft of Product is not just acceleration, but orchestration, that is knowing where to apply AI, where to defer to agents, and where to differentiate through uniquely human value.

AI began as an accelerator: faster research, faster prototypes, faster builds. It is now evolving into a market participant, capable of making and executing decisions on behalf of customers.

• Acceleration → Autonomy.

• Subsidy of inefficiency → Transparent optimisation.

• Human-only design → Dual-facing design (humans + agents).

• Speed → Focused restraint.

For Financial Services, this is beginning to look like an existential shift. Margins built on opacity will collapse, distribution becomes more algorithmic and Loyalty could be redefined.

Okay.. (I know I said no more) but it’s not all threat :

The opportunity, is equally profound. Greater focused productivity - those who design for both humans and agents, who own the right control points, and who practice this focused restraint in what they build, could shape the next era of both Product Management and finance.