Skip to main content

What we have learned, dated and sourced.

Field notes from running AI in our own products, the history behind how businesses answer their customers, and plain guides. Every article shows when it was published and where its facts come from.

Articles.

  1. Guide

    · 5 min

    Five questions to ask any AI product at Web Summit

    Web Summit Lisbon runs from 9 to 12 November 2026. Five questions that separate an AI product you can trust in front of customers from a good demo, and how to judge the answers.

    Read
  2. Analysis

    · 6 min

    AI agents and the ICO: what a UK business needs in place

    The ICO says an AI agent's autonomy is no excuse for poor compliance. What it expects from businesses using agents, the incidents behind it, and what to do now.

    Read
  3. News

    · 3 min

    Slipmark Cloud is live

    Slipmark Cloud, the hosted layer for the open-source Slipmark gate, is live at app.slipmark.dev. It keeps a publicly anchored record of every AI agent verdict, enforces review rules as a GitHub check and produces audit evidence.

    Read
  4. Analysis

    · 6 min

    Is AI eating entry-level jobs? What the UK figures say

    James Reed says AI is burning through entry-level jobs. What the UK figures show, what AI gets right and wrong, and why firms using AI should hire juniors.

    Read
  5. Analysis

    · 6 min

    AI found a decade-old flaw in PHP. Now what?

    Two critical PHP flaws, found with AI, sent sites across Europe offline in October 2026. What happened, and three lessons for a UK business that runs a website.

    Read
  6. Analysis

    · 6 min

    The AI hiring chatbot, a forgotten test account and the password 123456

    In 2025 researchers reached McDonald's McHire applicant data through a vendor test account with the password 123456. What failed, and the checks that catch it.

    Read
  7. Analysis

    · 6 min

    Fake cases, real consequences: the High Court on AI-invented citations

    In 2025 the High Court dealt with lawyers who filed case citations that did not exist. What the court decided, and the checks any business using AI needs.

    Read
  8. Analysis

    · 7 min

    When AI agents act on their own: Replit's deleted database and Cursor's invented policy

    Replit's AI agent deleted a production database in a code freeze; Cursor's support bot invented a policy. What happened in 2025 and the controls that fix it.

    Read
  9. News

    · 3 min

    Slipmark is live and open source

    Slipmark, formerly DriftGuard, is released as an open-source GitHub Action under Apache-2.0. It makes AI agents prove what they checked before anything ships.

    Read
  10. News

    · 3 min

    ReplyOS at Web Summit Lisbon 2026

    ReplyOS is exhibiting at Web Summit Lisbon on Wednesday 11 November 2026, booth A6-13. What we are bringing and how to meet the founder.

    Read
  11. News

    · 4 min

    What we build and operate: seven products, one standard

    ReplyOS, ReplyBench, ReplyScore, Slipmark, Weft, VetReady and Larkabout: what each product does, who it is for, its status and published prices, from Reply Intelligence Group Ltd.

    Read
  12. Analysis

    · 8 min

    When customer-facing AI goes wrong: five real cases and the control each one lacked

    Air Canada's chatbot ruling, DPD's swearing bot, the $1 Chevy Tahoe, NYC's MyCity and Snap's My AI: what happened, and the control that would have caught it.

    Read
  13. Analysis

    · 6 min

    Klarna's AI assistant: the claims, the headlines and what happened next

    In 2024 Klarna said its AI assistant did the work of 700 agents; in 2025 it stressed human support. What Klarna said, what headlines added, and the lessons.

    Read
  14. Field notes

    · 5 min

    The security gate that reported green for months

    A security gate in our own nightly AI agent pipeline inspected the wrong branch for months while reporting success. What happened, and the control we built.

    Read
  15. History

    · 7 min

    From switchboard to AI assistant: a short history of answering the customer

    Telephone exchanges, operators, answering machines, email, web forms, live chat and AI: 150 years of answering customer enquiries, dated and sourced.

    Read
  16. History

    · 6 min

    Speed to lead: what the research actually says

    The famous “21 times” figure is from a 2007 vendor study; HBR reported “nearly seven times” in 2011. What was measured, its limits, what holds.

    Read
  17. Guide

    · 7 min

    Customer-facing AI in a UK business: the rules that apply in 2026

    No AI-specific law applies in the UK. UK GDPR, as amended by the Data (Use and Access) Act 2025, does. What changed in 2026, plus a practical checklist.

    Read