An AI agent does not need malicious intent to create an expensive problem. It only needs access to something important and permission to take the wrong action.
That question is becoming urgent as agents move beyond generating answers. These systems can now access accounts, send emails, update records, communicate with customers, and complete transactions.
JPMorgan Chase has said it will restrict outside agents from entering customer accounts because the risks remain too significant.
Charles Schwab is also blocking unauthorised agents attempting to operate inside client accounts or move money. Instead, Schwab is developing controlled connections that could provide read-only access for checking balances and conducting research.
Financial services provide the clearest example because one incorrect action can create an immediate loss. However, the same issue affects every business connecting artificial intelligence with email, customer data, financial systems, supplier portals, or internal documents.
Capability alone should never determine an agent’s authority. Businesses need a permission ladder covering read-only access, recommendations, drafted actions, limited execution, and sensitive actions requiring human approval.
Every agent also needs a named owner, defined purpose, activity record, and immediate shutdown process.
What This Means For SMEs
SMEs should create a simple register covering every agent connected to company systems. Record what each agent can access, which actions it can complete, who approves sensitive decisions, and how access can be withdrawn.
Begin with read-only access wherever possible, especially across financial accounts, customer information, and supplier systems. Wider permissions should follow only after the workflow performs reliably under supervision.
What This Means For Enterprises
Enterprises need separate identities for agents alongside existing employee access controls. Security teams should establish permission limits, approval thresholds, audit logs, incident procedures, and regular access reviews.
Procurement and legal teams should also establish where responsibility sits when an external agent causes financial, operational, or reputational damage.
The strongest first movers will expand agent access without losing accountability. Before connecting another agent, decide what it may observe, what it may recommend, and what it must never complete alone.
Read the complete Reuters report and The Next Web analysis for further details.
The Briefing: What Else You Need to Know
Anthropic’s Junior Technical Hiring Falls From 27% To 9%
Anthropic’s internal workforce report shows junior employees falling from 21 per cent to 14 per cent across technical departments. Their share of technical hiring also declined from 27 per cent to 9 per cent, while employee-directed agents logged six hours for every assumed human working hour by August.
For SMEs, the warning concerns future capability rather than immediate redundancies. Businesses reducing junior recruitment should redesign those roles around quality assurance, customer understanding, agent supervision, and practical judgement, ensuring tomorrow’s experienced employees still have somewhere to begin.
Read Anthropic’s complete workforce report for further findings.
Microsoft Gives AI Agents Their Own Workplace Identity
Microsoft now allows Copilot Studio agents to receive their own Microsoft Entra identity, mailbox, and presence across Microsoft 365. Employees can interact with these agents through supported email, Teams, and Office scenarios, making them resemble digital colleagues working across shared processes.
Smaller businesses using Microsoft 365 should create an agent register before activating these capabilities. Every agent should have a named owner, defined purpose, controlled permissions, visible activity history, and documented shutdown procedure.
Review Microsoft’s agent account guidance before creating workplace identities.
AI Shopping Transactions Increase Thirty-Eightfold
Stripe data shared with Upstarts Media suggests transactions completed through Link-enabled agents increased thirty-eightfold during September, while payment volume increased twentyfold. Retail represented the leading category, although the figures began from a small group of participating agents.
Ecommerce SMEs should ensure their product descriptions, prices, availability, delivery terms, and returns information remain accurate and machine-readable.
Agents can only recommend and purchase products using the information they can reliably access. Explore the complete agentic purchasing analysis for additional context.
ChatGPT Advertisers Can Target Customer Lists And Conversation Context
A 404 Media investigation found that ChatGPT advertisers could create audiences using uploaded email addresses, telephone numbers, device identifiers, and signals from customer conversations. OpenAI reportedly prevents advertisers from identifying individual matches, although the development still introduces important consent and privacy questions.
SMEs should review their consent wording before uploading customer information to emerging advertising platforms. Businesses should also ask marketing agencies which data gets used, where it gets processed, and whether the activity has an appropriate lawful basis. Read the complete 404 Media investigation for further information.
AI Code Review Found More Problems Than Developers Could Fix
The Register reports that PostgreSQL 19 moved towards a November release after AI-assisted reviews surfaced numerous late-stage defects. Several planned features were reportedly removed because maintainers lacked sufficient time to investigate and resolve every problem before release.
For SMEs developing software, the lesson concerns capacity rather than coding speed alone. AI can identify weaknesses faster, although somebody must still assess severity, correct the code, verify every change, and decide what prevents release. Read The Register’s complete report for the technical background.
Nscale’s Valuation Meets Britain’s Infrastructure Reality
British AI infrastructure company Nscale is facing investor resistance over its proposed listing valuation. Microsoft has also withdrawn from a planned capacity agreement at Nscale’s Essex site after electricity-grid delays reportedly pushed the expected connection beyond 2030. Nscale continues discussing substantial computing agreements elsewhere.
SMEs purchasing services through smaller cloud providers should examine financial stability, infrastructure dependencies, data portability, service continuity, and exit arrangements.
AI services still depend upon physical facilities, reliable electricity, considerable financing, and suppliers capable of delivering promised capacity. Read the Financial Times report and Reuters coverage for further details.
The Shift - Your Personal Brand Is Becoming Your AI Visibility Strategy
Within an AI-driven search environment, your personal brand influences more than the people following your content. It also creates public evidence that helps AI platforms understand your expertise, relevance, and authority.
AI systems encounter information across websites, articles, videos, podcasts, social platforms, reviews, and third-party mentions. Publishing more content achieves very little when those signals create an inconsistent or forgettable picture.
1. Build Proof Worth Referencing
Strong visibility begins with genuine experience, measurable results, original research, useful frameworks, and informed opinions. These assets give customers something valuable while providing AI platforms with clearer evidence of your expertise.
Founders can document client outcomes, recurring customer questions, industry experiments, operational lessons, and decisions shaped by genuine experience.
2. Create A Recognisable Association
Your name should repeatedly appear alongside several relevant problems, services, and areas of expertise. Consistency helps customers remember your positioning while making your specialist knowledge easier for AI platforms to interpret.
Choose the commercial questions you want to become known for, then answer them consistently across your website, LinkedIn content, videos, podcasts, and published articles.
3. Strengthen Evidence Beyond Your Platforms
Claims published exclusively through your own channels carry limited independent authority. Podcast appearances, reputable publications, customer reviews, industry collaborations, and third-party citations create stronger supporting evidence.
Every appearance should reinforce the same expertise while contributing something genuinely useful to that audience.
First-Mover Action
Ask several AI platforms five questions that potential customers regularly ask before purchasing your service. Record whether your name appears, which competitors receive mentions, and which sources influence every answer.
Those results will reveal where your positioning remains unclear and where stronger evidence should be created.
The Big Question
If a customer asked an AI assistant who understands their problem best, would the available evidence provide a compelling reason to recommend you?
What We’re Working On
1. Building An Evidence-Led AI Visibility Audit
We are currently developing a more evidence-led approach to AI Visibility Audits. The process will explain what shapes a company’s visibility, rather than simply recording whether it appears within several prompts.
You will receive practical evidence showing where your visibility remains weak, why competitors may be performing better, and which improvements deserve immediate attention.
Want to understand whether AI recommends your business, which competitors appear instead, and what may be influencing those results? Reply “LIVE AUDIT” to get your AI Visibility Audit done.
2. FDE Support Is Available For UK Businesses
The main story explains why businesses need somebody responsible for turning AI opportunities into practical improvements. The GATE Project provides an accessible way for SMEs to begin that work without immediately recruiting a six-figure AI leader.
Each business receives a Future Development Engineer for twenty weekly hours across a structured three-month placement. Associates combine software-engineering experience with additional training across process automation, AI agents, n8n, Claude operating systems, and business implementation.
For £600 monthly, businesses receive 240 hours of support identifying repetitive processes, building initial automations, documenting internal knowledge, and improving how work moves across teams.
To discuss The GATE Project or request an intern, email [email protected] or call 0121 517 2258.
3. Brand Collaborations And Media Partnerships
We believe every business increasingly needs to operate like a media company within today’s interest-led and AI-driven landscape. That belief also shapes how we approach our own growth, with media sitting at the centre of our strategy.
Brands including Riverside, Descript, and Music Creator AI have approached us about creating videos for our YouTube and social channels. These opportunities allow us to introduce useful platforms through practical content that genuinely serves our audience.
If your organisation would like to explore a media collaboration with us, email [email protected] to request our partnership options and pricing.
4. Podcast Conversations With Dean Whitby
Our founder, Dean Whitby, is currently looking for podcasts and media platforms where he can share practical insights about AI implementation.
Dean specialises in AI visibility, workflow integration, process automation, and helping businesses apply artificial intelligence within genuine operations. His conversations focus on what leaders can implement, measure, and improve within their organisations.
If you host a podcast, interview series, YouTube channel, or business platform, email [email protected] to discuss hosting Dean for an upcoming conversation.
Final Takeaway
AI is increasingly acting inside businesses while influencing which businesses customers discover outside them. That shift creates two responsibilities that leaders cannot afford to separate.
Internally, businesses need clear permissions, named ownership, human approval points, and reliable activity records. Externally, they need consistent evidence demonstrating their expertise across websites, social platforms, videos, publications, reviews, and trusted third-party sources.
The strongest first movers will combine responsible implementation with credible visibility. They will understand what their agents can access, while ensuring AI platforms can accurately understand what their business offers and why customers should trust it.
Choose one connected agent and review every permission it currently holds. Then ask several AI platforms five questions your customers regularly ask before making a purchase.
Together, those exercises will reveal whether artificial intelligence represents a controlled advantage, an unmanaged risk, or a missed commercial opportunity within your business.
Want to Know What AI Is Saying About Your Business?
If you want to understand how visible your business is in AI search and how AI compares you against your competitors, we can help.
Reply “live audit” to this newsletter, or email [email protected], and our team will run an AI Visibility Audit so you know exactly what AI is saying about your business.
Is this newsletter right for you?
We send this briefing to help first movers stay close to the AI news, trends, and signals that matter.
If that is useful to you, we would love to keep sharing it with you. If it is not quite what you are looking for, no problem at all, you can unsubscribe using the link below.
Either way, thank you for giving it a read.
