Isomorphic Labs is lining up more than £2 billion in funding for AI drug development in the UK. This raise could cut discovery timelines and redraw investment maps. Watch closely.
OpenAI and the United Kingdom have a new playing field for advertisers with the self-serve Ads Manager for ChatGPT. This product brings CPC bidding and conversion tracking to marketers and promises sharper ad performance—read on for what to do next.
OpenAI and the UK collide over a $30bn stake disclosure that shifts trust and governance priorities. This matters to UK executives and investors. The fallout will test boardroom scrutiny and regulatory resolve.
AI models in the United Kingdom are steering search results and rewriting the rules for visibility. Understand the 'black box' and you can turn mystery into market share.
<p>Samsung Galaxy users in the UK can now access Google Gemini AI Assistant inside Samsung apps, changing how phones help you daily.</p><p>This update tightens app workflows and opens new commercial pathways for product teams and mobile developers. A decisive step for smartphone AI.</p>
Nebius acquisition lands squarely in the UK conversation about AI infrastructure and competition. The deal forces enterprise technology teams to rethink model optimisation and vendor strategy — fast.
Serverless hosting in the UK just gained a loud new voice as Featherless.ai raised $20 million. The funding will expand global infrastructure and launch an AI marketplace for open-source AI models. Democratisation, meet infrastructure.
Rogo Technologies in the United Kingdom has raised $160 million to speed financial analysis with AI. The investment accelerates tools that make financial professionals far more productive; the race is on.
Parag Agrawal in the UK has seeded a new platform to let AI agents search the web more like a human expert. The move could rewire how businesses retrieve data and build products; consider this your wake-up call.
QCraft unveiled a physical AI model at the Beijing auto show in China, signalling a leap in autonomous vehicles and automotive technology. The demonstration is a bellwether for manufacturers and suppliers; act now.
In the UK, AI governance has leapt to the front of defence and procurement debates after Anthropic denied Pentagon claims about a 'kill switch'. This legal challenge forces firms and regulators to choose clarity over assumptions.
Cohere in the United Kingdom has announced a merger with Aleph Alpha backed by $600 million from Schwarz Group. The deal signals rapid consolidation and fresh scale for retail-focused AI. Expect fast consequences.
DeepSeek and the UK now face fresh market questions after the company unveiled a new flagship AI model. This development forces investors and tech leaders to get strategic, fast.
In July 2025, Brussels reaffirmed that the Digital Services Act (DSA) and the Digital Markets Act (DMA) were non-negotiable in trade talks with Washington. Twelve months later, that posture reads as a warning shot. The EU has issued its first major DSA fine, put Meta and TikTok on preliminary notice, opened a formal investigation into an AI chatbot, and published the enforcement architecture that runs through August 2026. For any Very Large Online Platform — or any brand advertising, moderating content, or shipping AI features into Europe — the wait-and-see window has closed.
Claude Sonnet 4.6 shipped on 17 February 2026 and, two months on, the numbers have settled. It scores 79.6% on SWE-bench Verified — 1.2 points behind Opus 4.6 at a fifth of the price — carries a 1-million-token context window as standard, and powers the fastest-growing enterprise software product in history. Claude Code crossed £2.5B in annualised run-rate revenue in February 2026, doubling in eight weeks. If you lead an engineering team in the UK, the question is no longer whether to adopt Sonnet 4.6 for coding — it's how fast you can wire it into the way your team ships. This guide replaces the speculation that surrounded the 4.6 launch with verified 2026 data: benchmarks, UK adoption, pricing, and a five-step operational rollout.
When this post first went live in July 2025, Meta's 'proactive chatbot' ambitions were still a leak — a reported AI Studio experiment to nudge purchases after five interactions. Nine months later, that leak has become the floor of a much bigger build: Meta is rolling AI shopping agents across Instagram, Facebook and WhatsApp; it has launched a dedicated shopping model called Muse Spark out of Meta Superintelligence Labs; and as of 15 January 2026, WhatsApp has formally banned general-purpose chatbots while keeping the door open for structured, purpose-built ones. The ground under proactive shopper engagement has moved — and most of the 2025 tactical advice no longer matches the platform.
In 2025 we argued that ethical AI was a competitive advantage. In 2026, that framing is too soft — trust is now the last moat most marketing brands have, and AI is eroding it for everyone who handles it sloppily. Three things have converged: the EU AI Act's transparency obligations land in August, New York's SB S8420A forces AI disclosure in advertising from June, and the IAB has put a 37-point number on the gap between how advertisers think Gen Z feels about AI ads and how Gen Z actually feels. The brands still growing this year are the ones consumers can believe.
When we first published this piece in late November 2025, Claude Opus 4.5 was the headline. Five months later that framing is obsolete. Anthropic has shipped two further Opus releases — 4.6 in February 2026 and 4.7 on 16 April 2026 — and the gap between teams running the current model and teams still quoting 4.5 benchmarks in sales decks is now the real competitive edge. If you are a UK founder, agency owner or in-house marketer trying to work out what to actually do in the week after the 4.7 launch, this is the version that matters.
Cloudflare didn't just draw a line in the digital sand in 2025. By April 2026, that line is a wall — 3.8 million domains thick, policed by a permission-based crawler economy, and reshaping how AI companies get their training data. Over 1 million customers took the aggressive block option in year one, 89.4% of AI crawler traffic now serves training rather than search, and ClaudeBot still crawls roughly 73,000 pages for every referral it sends back. If you publish content and haven't thought about the blockade, you're either being scraped without compensation or you've been cut out of the LLM answers where buyers now live. There is no neutral position.
If you sell anything in the United Kingdom and Google is still your main traffic source, the last five months have quietly rewired your funnel. Gemini 3 is now the default model behind AI Overviews globally. Gemini 3.1 Pro shipped into AI Mode on 22 April 2026. AI Mode itself has rolled into nearly 120 countries, with UK-specific agentic features going live in the same window. And Apple has confirmed that Siri — the assistant sitting in roughly half of UK pockets — will be running on Google's Gemini from iOS 27 onwards. This is not a new feature drop. It is a change in who owns the first answer a British buyer sees.
For nearly a year, Google AI Max for Search sat in an open beta that hundreds of thousands of advertisers dipped into cautiously — some saw double-digit lift, some saw budget evaporate into zero-conversion search terms, most sat on the fence. That fence is now gone. On 15 April 2026, Google announced that AI Max for Search has reached general availability, and that Dynamic Search Ads, Automatically Created Assets and the campaign-level broad match setting will all auto-upgrade into AI Max by September. If you run paid search, the keyword-first playbook you learned in 2010 officially stopped being the default this month. This is the rewrite of that story — what shipped, what's actually true about the numbers, and what to do before Google makes the decision for you.
On 12 February 2026 at 21:12 UTC, Supabase's us-east-2 region in Ohio went dark for three hours and forty-two minutes. Database connections failed, API endpoints timed out, and roughly 4.92% of Supabase's customer base lost access to production. Twelve days later, 120,000+ Indian developers were cut off from Supabase entirely for a 7–8 day network-level blackout. If your 2026 SEO strategy doesn't assume your backend will fail, it's not a strategy — it's a wish.
For eighteen months, the story was simple: Microsoft bet on OpenAI, Google bet on Gemini, Amazon propped up Anthropic. In 2026 that story is dead. Claude is now the default model inside Microsoft 365 Copilot for most commercial tenants, Copilot Cowork is the flagship feature of the new Microsoft 365 E7 tier, and Claude Opus 4.6 sits in Microsoft Foundry alongside GPT. If you're running a marketing team on Copilot and you don't understand what changed, you're getting outputs from a different model than you think you are.
In July 2025 we wrote about an 'OpenAI super-browser' rumoured to arrive within weeks. It did — and the reality is bigger and messier than the leak suggested. ChatGPT Atlas launched on 21 October 2025, Operator was folded into Agent mode, and the competitive field now includes Perplexity's Comet, Atlassian-owned Dia, Brave Leo and Opera Neon. Nine months on, the browser is the new front door to the web — and most marketing teams are still optimising for the old one.
In January 2026, a Scientific Reports study compared more than 100,000 humans against frontier language models on divergent-creativity tasks. AI now beats the average person — but the top 10% of humans still leave AI behind, and the gap widens on longer-form creative work. The real story of creativity in 2026 isn't that AI replaces you. It's that there is now a measurable line between people who use AI to stay average and people who use AI to get further ahead. Which side of that line you sit on is almost entirely a function of how you collaborate with the model, not which model you use.
In July 2025 we argued Perplexity was about to fire a shot at Google's advertising empire with sponsored follow-up questions priced at a fraction of AdWords CPMs. Nine months later the exact opposite has happened. Perplexity pulled its ad product, pivoted to a subscription-first model and is chasing $500M in annualised subscription revenue. Google, meanwhile, quietly shipped the biggest change to Search ads in a decade: AI Max left beta on 15 April 2026 and will automatically swallow every remaining Dynamic Search Ads campaign in September. The ad-model race didn't end in a duel — it ended with one competitor walking off the field and the other rewriting the rules.
In late August 2025, Meta filed paperwork in Sacramento for a California super-PAC with an unusually on-the-nose name: Mobilizing Economic Transformation Across California. Eight months on, the story is no longer a Politico scoop — it is a $65M political infrastructure spanning four PACs, two freshly signed California AI laws, and a March 2026 executive order from Governor Newsom that pulls AI vendor policy into state procurement. If you run marketing in 2026, the politics of AI in California now shape which tools you're allowed to use, what you must disclose, and how you buy. This piece is the refreshed, post-signature version — what the PAC actually spent, what the laws actually require, and where marketers get caught.
When Databricks announced its $1 billion acquisition of Neon in May 2025, most coverage framed it as another big-cheque AI data deal. A year later, the framing looks small. In February 2026, Databricks shipped Lakebase — a fully managed, Neon-powered serverless Postgres service — into General Availability on AWS, followed by Azure GA at FabCon 2026. Databricks has disclosed that more than 80% of new databases on the platform are now provisioned by AI agents, not humans. If you are building anything where an AI system reads, writes, branches or reasons over structured data, this is the infrastructure story of 2026.
Google DeepMind's AlphaEarth Foundations is the closest thing the public internet has to a working virtual satellite. It fuses optical imagery, radar, LiDAR and climate simulations into a single 64-dimensional embedding for every 10×10-metre patch of land and shallow water on Earth — and as of March 2026, the 2025 state-of-the-planet embeddings are live in Google Earth Engine, Google Cloud Storage and BigQuery. Error rates are 23.9% lower than the next-best Earth observation models. Storage footprint is 16× smaller. For anyone working on climate risk, supply-chain traceability or sustainability reporting, this is the most important free dataset on the internet right now.
On 3 December 2025, OpenAI announced it had agreed to acquire Neptune.ai, the Polish experiment-tracking platform, for an all-stock deal reported at under $400 million. Three months later, on 5 March 2026 at 10am PST, Neptune's hosted app and API went dark and customer data was irreversibly deleted. This was not a quiet rollup — it was OpenAI absorbing a category leader, shutting down the external product, and walking away with the people and the IP. If you were running Neptune in production in December, you had 92 days to export, migrate and rebuild your ML experiment-tracking stack from scratch. The tweetable version is 'OpenAI strengthens training infrastructure.' The real story is that a thousand ML teams had a vendor shutdown forced on them during the busiest shipping quarter of the year.
AI funding platforms are no longer a pitch — they are the default way capital finds founders in the United States. In Q1 2026, investors deployed $300 billion into 6,000 startups globally, with roughly $242 billion (80% of total VC) going to AI companies. Four rounds alone — OpenAI, Anthropic, xAI and Waymo — soaked up 65% of that capital. If you are a US founder in 2026, your fundraising outcome is increasingly decided by how well you show up inside three or four matching engines, not by how many warm intros you can scrape out of LinkedIn.
Updated 23 April 2026. The original October 2025 version of this piece treated OpenAI's adult-content announcement as the story. It wasn't — it was the starting gun. Since then OpenAI has published a Teen Safety Blueprint, delayed Adult Mode twice, stripped it down to text-only, and watched the UK quietly fold AI chatbots into the Online Safety Act with Ofcom fines of up to 10% of global turnover. If you deploy ChatGPT anywhere near UK consumers, the operating reality in April 2026 is not the one the original post described. This is the refreshed playbook.
On 12 March 2026, a fourteen-month-old San Francisco startup called Axiom announced a $200M Series A at a $1.6B valuation, led by Menlo Ventures. The thesis is narrow and unusually bold: if AI is going to write most of the world's code, something other than another LLM needs to prove that code is correct. Axiom's answer is a Lean-based proof engine that treats verification as a mathematical problem, not a linting one. For every marketing, product and agency team now shipping AI-generated software into production, this is the week the 'trust' conversation moved from vibes to formal proof.
When we first published this piece in May 2025, the question for marketers was whether to trust generative AI at all. Twelve months and three flagship model releases later, that question is settled. The live question is far more interesting, and far more expensive to get wrong: which AI model do you route which piece of marketing work to — and how fast can you switch when the leaderboard changes? Gemini 3.1 Pro shipped in February 2026. GPT-5.4 followed in early March. On 16 April, Anthropic released Claude Opus 4.7 and narrowly retook the benchmark lead. Your 'AI brain' isn't one thing any more — it's a routing decision you make dozens of times a day.
In May 2025, Sir Elton John called the UK government's plan to let AI firms train on copyrighted work without consent a “criminal act.” Ten months later, on 18 March 2026, Technology Secretary Liz Kendall stood up in Parliament and effectively agreed. The opt-out exception — the one Elton had attacked on BBC, the one Paul McCartney, Kate Bush, Dua Lipa and 400 other artists had signed open letters against, the one that forced 1,000 musicians to release a silent album in protest — is no longer the government's preferred way forward. If you write a brief that relies on AI-generated assets, train a model on client data, or run marketing at any kind of scale, the rules you were betting on have just changed. This is the refreshed, post-U-turn version of our original May 2025 explainer.
A year ago Grizzly New Marketing quietly launched an AI-adapted content optimisation service for a problem most agencies were still pretending wasn't happening. Twelve months on, the numbers end the argument. Seer Interactive's 2026 analysis of 3,119 informational queries shows organic CTR collapsed 61% — from 1.76% to 0.61% — on searches where Google shows an AI Overview. AI Overviews now trigger on ~48% of tracked queries and 82% of B2B technology searches. If your content strategy is still built around ten blue links, you aren't losing slowly — you're losing at ~60% per quarter.
For a year, marketers treated 'AI consumer behaviour' as a forecast. In 2026 it stopped being a forecast. ChatGPT has over 800 million weekly active users, 35% of initial brand discovery now happens inside an AI answer, and 64.82% of Google searches end without a single click. The consumer journey isn't starting at your homepage anymore — it's starting, and often ending, inside a conversation your brand was never invited to. This piece updates our original May 2025 analysis with what actually happened: the Agentic Commerce Protocol, the Gemini-over-Perplexity referral flip, the real state of Gen Z search, and the strategy changes marketers have to make right now.
Eight months ago, Google DeepMind unveiled Genie 3 as a research prototype — a general-purpose world model that turned text prompts into navigable 720p environments at 24 frames per second. It looked like a tech demo. In 2026 it is a shipped consumer product, a Waymo training tool, and the opening move in a world-models race that has already attracted more than $10 billion in venture funding. If you run marketing, the shift from 'AI that generates video' to 'AI that generates worlds you can walk around in' is not an abstract research story. It rewrites what a campaign, a product demo and a brand experience can be.
In May 2025, Blue Yonder announced Cognitive Solutions — a platform of autonomous AI agents designed to monitor, evaluate and respond to enterprise supply chain workflows in real time. Eleven months later, that platform has stopped being a deck and started being a product line. In March 2026, Blue Yonder expanded the suite with new industry-specific agents, a mobile command app called Orchestrator, and native Microsoft Teams integration. In April 2026, Gartner named Blue Yonder a Leader in the inaugural Magic Quadrant for Supply Chain Planning Solutions (Discrete Industries). If you want to understand what 'agentic AI' actually looks like inside a Fortune 500 operator, this is the case study.
On 22 January 2026 the team behind open-source vLLM walked out of UC Berkeley's Sky Computing Lab with $150 million in seed funding and an $800 million valuation. Three months later, the UK government announced a £500 million Sovereign AI Fund whose first cohort includes Doubleword — a British self-hosted inference platform. These two events are the same story viewed from opposite ends of the Atlantic: the plumbing of AI is being commercialised, and inference cost is now the single biggest lever in enterprise AI economics. If you run AI workloads in the UK and you're still paying per-token to a US frontier API, you are now paying the most expensive price available in a market that just gained three credible alternatives.
In October 2024, DeepMind's Demis Hassabis and John Jumper shared the Nobel Prize in Chemistry with the University of Washington's David Baker — AlphaFold for protein structure prediction, Baker's Rosetta for computational protein design. Eighteen months later, that acclaim has compounded into something very different from a polite research footnote: a proprietary drug-design engine called IsoDDE, roughly $3 billion in contracted milestone payments from Eli Lilly and Novartis, and the first AI-designed cancer drug heading into Phase 1 clinical trials this year. The Nobel was the starting gun. The real story is what DeepMind did with the next 18 months.
When Palo Alto Networks first signalled its intent to buy Protect AI in April 2025, it was a single line item in a longer cyber-security consolidation thesis. Eleven months on, the picture is very different. The deal has closed, Protect AI's model-scanning and red-teaming engines now sit inside Prisma AIRS 3.0, Palo Alto has shipped a full agent-workforce platform called Cortex AgentiX, and the wider competitive board has been redrawn by Alphabet's $32 billion Wiz close and Palo Alto's own $25 billion CyberArk acquisition. If you're a CISO, a platform lead, or a marketer whose AI stack is quietly piling up agents, this is the moment to re-read the map.
When we first published this piece in May 2025, the conversation was still theoretical — would agentic AI replace jobs, augment them, or rewrite the org chart? Eleven months later the question has moved. Agentic systems that plan, execute and self-correct across multi-step workflows are now live inside IBM, Klarna, Salesforce, BT, JPMorgan and most of the Fortune 500. The org charts those companies publish internally no longer resemble their 2023 versions. This April 2026 update tracks what changed: the pyramid is cracking, the middle is compressing, and redesigning employment models around human-agent teams has become an HR quarter-end deliverable.
ClaimSorted just closed a $13.3m seed round led by Atomico, with Eurazeo and Y Combinator writing cheques alongside. In the same six months, Aviva disclosed £60m+ in annual savings from 80 AI models inside its claims function, and the FCA told the market that claims handling is a top-tier supervisory priority for 2026. If you run marketing, pricing or ops inside a UK insurer, broker, MGA or insurtech, the competitive map has been redrawn — and most teams haven't updated their plan. This refresh pulls the 2026 developments into one place: what ClaimSorted actually built, what the FCA is now demanding, what carriers like Aviva have proved is achievable, and how to translate all of it into commercial uplift you can sell internally.
When Truth Social quietly switched on an AI search engine in August 2025, the reactions split into two camps. One side called it a political stunt. The other called it a Perplexity distribution play. Both missed the more interesting story: a social platform built around one man's narrative bought an AI that refuses to agree with him — and then tried to contain the damage by controlling which websites it reads from. Eight months on, the picture is larger and stranger. Truth Search AI is still live, Trump Media is mid-way through a ~$6 billion merger with TAE Technologies, Devin Nunes has stepped down as CEO, and DJT stock is down more than 30% year-to-date. For marketers, this is no longer a curiosity — it's a case study in what happens when AI search collides with brand control.
Between November 2025 and April 2026, UK retail's front door quietly moved. It is no longer google.co.uk and it is no longer the high street — it is the answer box inside ChatGPT, the chat pane inside Perplexity, and Google's AI Mode tab. Shopify's Agentic Storefronts went live by default on 24 March 2026, PayPal + Perplexity Instant Buy is shipping UK orders via Debenhams, and John Lewis has committed its £800m transformation programme to Gemini and ChatGPT. If your products aren't discoverable there, you are invisible to a growing slice of UK demand that hasn't yet shown up in your analytics.
In 2025 'ethical AI marketing' was a panel topic. In 2026 it's a line item with a deadline. On 2 August 2026 the transparency obligations of the EU AI Act become enforceable — covering any brand, agency or in-house team that publishes AI-generated text, image, audio or video to EU audiences. Penalties run up to €35 million or 7% of worldwide turnover. The FTC closed its second AI-washing case in February 2026 and New York's synthetic-performer disclosure rule lands in June. This piece is the operational version of ethical AI marketing: what's now legally required, what's being enforced, and what to change before the August deadline.
On 3 December 2025, Anthropic announced its first-ever acquisition: Bun, the all-in-one JavaScript runtime, bundler, package manager and test runner written in Zig. The deal landed the same week Claude Code crossed $1 billion in annualised revenue, six months after launch. Four months on, Bun is no longer an external dependency — it's the runtime under Claude Code, under the Claude Agent SDK, and under the next generation of AI coding products Anthropic is quietly shipping. If you're a founder, agency owner or marketing leader wondering whether this is engineer news or a shift that touches you, it's the second one — and this post explains why.
For most of 2025, "AI writes code" meant autocomplete with a chat interface. In April 2026, Google DeepMind quietly moved the goalposts — setting AlphaEvolve loose on the Python source of game theory algorithms and watching it rewrite them into two new variants, VAD-CFR and SHOR-PSRO, that beat every human-designed baseline. This is not AI helping a researcher. This is an AI that edits the algorithm, runs it, scores the result, edits it again, and keeps going until it beats the expert. Applied to maths, it has improved a lower bound that has stood since the 1970s. Applied to Google's own data centres, it is clawing back efficiency gains worth hundreds of millions a year.
Five months ago, DeepSeek was a headline about a Chinese lab undercutting OpenAI on inference cost. Today it is a headline about the first open-weight model to score gold at the International Mathematical Olympiad — and the Department for Education has just put £3m on the table to put AI tutors in front of 450,000 disadvantaged pupils. The two stories are the same story, and UK education leaders have about one term to work out what they do with it. This post was originally published in late 2025, before DeepSeekMath-V2 existed, before DeepSeek R2 launched, and before the DfE formally opened its tutoring tender. The update below reflects where the state of play actually is in April 2026.
When Perplexity announced a $42.5 million revenue-sharing pool for publishers in August 2025, most coverage stopped at the headline number. Eight months on, the programme has moved from press release to live product, Comet Plus has named launch partners, and Comet has shipped on every major platform. Meanwhile The New York Times and Dow Jones are still suing Perplexity in federal court while their peers cash the cheques. If you write content for a living — or advise brands that do — the question is no longer 'is this real?' but 'what share do you get, under what terms, and is it worth the legal exposure?' Here is a straight answer, updated for April 2026.
When we first wrote about Harvey AI in June 2025, the company was in talks to raise at a $5 billion valuation and the legal sector was still treating generative AI as an interesting pilot. Ten months later, Harvey is an $11 billion company with $190 million in annual recurring revenue, a majority of the AmLaw 100 as customers, and a strategic alliance with LexisNexis that reshapes how legal research will be bought and sold. The brief, as the original post put it, has now been filed — and Big Law is buying in bulk.
If you landed here searching for a '2025 LinkedIn algorithm update,' the honest answer is that the document you were looking for is already obsolete. Between June 2025 and April 2026, LinkedIn didn't tweak its ranking system — it replaced it. Velocity-based ranking is gone. Engagement pods are detected with 97% accuracy. The first hour still matters, but for a very different reason. This is the full timeline, what each change actually does, and the playbook that survives the new rules.
In July 2025 Cloudflare flipped a switch most marketers still haven't priced in: new domains block known AI crawlers by default, and a pay-per-crawl mechanism went into private beta. In January 2026 that beta moved to general availability, settling in-protocol over HTTP 402. February 2026 brought the Stack Overflow partnership, March 2026 shipped the native /crawl endpoint, and the New York Times and Reddit both sued Perplexity over alleged stealth crawling. If your SEO or PR strategy still assumes 'robots.txt plus a sitemap plus some backlinks' is the full picture, you're working off a 2023 map. This is the 2026 version.
When we first covered Apple's AI crisis in August 2025, the speculation was about which AI company Tim Cook would finally write a cheque for — Mistral, Perplexity or Anthropic. None of it happened. On 12 January 2026, Apple announced a multi-year partnership with Google for a custom 1.2 trillion parameter Gemini model, at a reported cost of roughly $1 billion per year. Apple didn't buy an AI lab — it rented one. This is the post updated for what actually happened, why the acquisitions fell apart, and what the 'rent, don't own' pattern means for any marketing team still trying to decide whether to build or integrate.
When Google I/O 2025 closed out on 21 May, most of the big reveals were promises — AI Mode was a rebrand, Gemini Agents were a slide, Project Mariner was a tab demo. It's now April 2026 and we finally know what those announcements meant. AI Mode crossed 75 million daily active users, Gemini 3.1 Pro shipped in February, Personal Intelligence rolled out to free-tier users in March, and the Gemini app passed 750 million users. AI Overviews now fire on roughly 48% of all Google queries. This is the I/O 2025 recap, updated for what actually shipped — and what it means for marketers heading into I/O 2026.
Tencent's Hunyuan line is no longer a single '2.0' model — it is a matrix. Since we first covered Hunyuan 2.0 in December 2025, Tencent has shipped Hunyuan-T1 (a Hybrid-Transformer-Mamba MoE scoring 87.2 on MMLU-Pro, second only to OpenAI o1), a visual reasoning sibling (T1-Vision), an end-to-end voice model (Hunyuan Voice), HunyuanVideo-1.5, and a 3D engine with 3M+ Hugging Face downloads. A new ~30B flagship led by ex-OpenAI researcher Shunyu Yao is scheduled for April 2026. If you're a UK marketing leader still thinking of Hunyuan as 'the Chinese MoE model', the map has been redrawn underneath you.
Nine months ago, we wrote about GPT-5 as a rumour — a leaked context window, an anonymous OpenAI engineer quoted in Decrypt, a summer launch that kept slipping. Today it reads like a time capsule. GPT-5.4 shipped on 5 March 2026 with a 1,050,000-token context window, a native Computer Use API, and $2.50 per million input tokens — pricing nobody in the 2025 rumour cycle saw coming. Rather than delete the old post, we've rewritten it as a retrospective: what the community got right, what we got wrong, and what a marketing team should actually do with a million-token context window now that the rumours are spec sheets.
Gartner predicts 40% of enterprise applications will feature task-specific AI agents by the end of 2026, up from less than 5% in 2025. Translation: every serious software vendor is now an 'agent platform' vendor, whether the badge fits or not. In the last ninety days alone, Box has shipped the unified Box Agent, NTT DATA has launched NVIDIA-powered AI Factories and confirmed an April Litron Builder release, Google has unveiled the Gemini Enterprise Agent Platform at Cloud Next, and AWS has pushed Bedrock AgentCore Policy to general availability. If you evaluated this space in 2025, your shortlist is already stale. This guide keeps the three platforms the original post profiled — NTT DATA, Box, and Nimanode — and folds in the 2026 hyperscaler wave that now sits on top of them.
OpenAI Grove is not an accelerator. It is a five-week technical mentorship at OpenAI's San Francisco HQ for people deeply curious about building in AI but with no startup yet — no pitch, no prototype, sometimes not even a co-founder. Cohort 1 ran in autumn 2025. Cohort 2 wrapped on 27 February 2026 with roughly 15 participants, $50,000 in API credits each and early access to unreleased models. If you missed both, the real question is not whether Grove is worth it — it is what you do between now and the next application window.
In April 2026 the BBC's audience complaints system quietly stopped being human-first. Serco's five-year, ~£40 million Audience Services contract went live — and with it, a layer of AI agents that now sort, cluster, triage and draft replies to the thousands of complaints the BBC receives each week. The headlines framed it as a procurement story. It isn't. It's the moment public-service complaint handling in the UK became an AI product category — weeks before a separate law makes a formal complaints procedure mandatory for every UK data controller. If you run customer operations, marketing, or any brand that regularly receives structured feedback, this is your template.
Tokyo-based Turing closed its Series A at ¥15.27bn (roughly $98.6m) at a $388m valuation in November 2025. Five months later the story is no longer the cheque — it is what the cheque bought. On 26 March 2026 Turing became the first company to drive a Vision-Language-Action model car on Japanese public roads in real time. On 17 April, Prime Minister Sanae Takaichi's Council for Japan's Growth Strategy formalised a national target: 25% of the global autonomous-vehicle market by the 2030s. If you sell to, market inside, or compete against any part of Japan's mobility stack, the ground you're standing on moved this spring.
When we first wrote this piece in July 2025, Llama 4 Scout and Maverick had just dropped and GPT-4 Turbo was still the model to beat. Nine months later, almost every sentence in that original post needs an update. Meta has shipped Llama 5 (600B+ parameters, agent-native, announced 8 April 2026). Behemoth is still in training but now casts a longer shadow. DeepSeek, Qwen and Zhipu's GLM family have collapsed the gap between open weights and frontier proprietary models. And Meta has pivoted hard — from 'let's have a chatbot' to 'let's own the agent layer of advertising.' This refresh tells you what changed, what stayed the same, and where the real commercial leverage is today.
When we first covered LinkedIn's push against pod spam in August 2025, the platform was hinting that an AI-driven crackdown was coming. It arrived — and then some. In early 2026 LinkedIn pulled its entire legacy ranking system and replaced it with 360Brew, a single 150-billion-parameter foundation model. Pod detection now runs at 97% accuracy, flagged accounts watch reach collapse from ~8,500 impressions per post to ~340, and recovery takes 60–90 days with no warning and no appeal. This post is about the surge itself: why LinkedIn's detection got this good this fast, how it actually works under the hood, and what it signals for every other platform still pretending not to notice coordinated engagement.
Salesforce built the modern CRM. In 2026 it's trying to build the operating system for every enterprise worker, human and AI. Eighteen months after the original Agentforce launch, Marc Benioff has something he didn't have in 2024 — a paying customer base, a named product line (Agentforce 360), and a credible $800M ARR story. This piece is a reset: what Agentforce actually does now, how it stacks up against Microsoft Copilot and Google Gemini Enterprise, and what the agentic enterprise means if you run marketing rather than IT.
Meta spent a decade and tens of billions of dollars telling us the future was VR. In early 2026, the company quietly conceded that the future is mobile, AI-generated and built by creators who have never touched a headset. Horizon Worlds — the flagship of Mark Zuckerberg's metaverse bet — is being reshaped in front of our eyes, and the tools Meta is shipping for creators are unrecognisable from the ones we covered when this post first went live. This is the refreshed, 2026 picture of Horizon, Horizon Studio, the new Horizon Engine, and the GenAI creator stack — plus an honest read on what it means if your brand was planning any kind of Horizon strategy.
Google AI Overviews were interesting in 2024, alarming in 2025, and — as of the January 2026 data and the March 2026 Core Update — the defining force in organic search. AIOs now appear in 25.8% of all US searches, 39.4% of informational queries and a staggering 82% of B2B Technology queries. The most rigorous study to date measured a 46.7% relative decline in organic CTR when an AIO is present, with position-one clicks collapsing from 28.5% to 11.2%. If you are still running the 2023 SEO playbook, your traffic is being quietly decommissioned. This is the refreshed playbook for what actually works now, and a candid account of why we are rewriting every one of our old posts.
Six months ago the story was “ChatGPT is becoming a checkout.” On 4 March 2026, OpenAI quietly switched Instant Checkout off. Three weeks later, Shopify flipped a different switch — and suddenly every store on the platform, not thirty hand-picked brands, became discoverable inside ChatGPT. That is not a delay. That is a strategic reversal, and it changes what AI commerce actually means for the next decade.
When Perplexity and PayPal announced their agentic commerce partnership on 14 May 2025, most commentary treated it as another AI-commerce press release. A year later, the aspiration has shipped. US users can type a shopping query into Perplexity, receive unbiased recommendations from 5,000+ merchants, and complete checkout without leaving the conversation — passkey authentication, tokenised wallet, tracking, returns and invoicing handled in the background. After OpenAI pulled back on Instant Checkout on 4 March 2026 and Shopify's late-March Agentic Commerce Protocol punted checkout back to merchant sites, Perplexity now owns the cleanest end-to-end 'search → buy' story in the market.
When Gartner crowned agentic AI the top strategic technology trend of 2025, most marketing and operations leaders nodded politely and carried on with their roadmaps. A year on, the nodding has stopped. Gartner's 2026 data has reframed agentic AI from 'interesting trend' to 'measurable inflection' — with 40% of enterprise applications expected to feature task-specific AI agents by end of 2026 (up from <5% in 2025), a $450 billion value projection by 2035, and a counter-signal that more than 40% of agentic AI projects will be cancelled by end of 2027. This is what the refreshed picture looks like — and what it means if you're the one signing the cheque.
When we first covered Hugging Face's Open Computer Agent last year, it was a scrappy demo — an open alternative to Operator and Claude computer use that clicked buttons, filled forms and fell over a lot. The headline in April 2026 is very different: the open community just won the benchmark. On 1 April 2026, Hcompany released Holo3-35B-A3B on Hugging Face, and within days it posted 78.9% on OSWorld-Verified — ahead of GPT-5.4 and Claude Opus 4.6, at roughly one-tenth the cost per run. The weights are open, the API is live, and a Chrome extension called HoloTab already puts a state-of-the-art computer-use agent inside any browser. That is the story — and everything in this post is the context around it.
If you're still running a LinkedIn micro content strategy built on 2024 playbooks, you're publishing into a feed that doesn't exist anymore. In early 2026 LinkedIn replaced its ranking system with 360Brew, a 150-billion-parameter model that scores every post on Depth Score rather than surface likes. Carousels with exactly seven slides now outperform other lengths by 18%, document posts are hitting a 6.60% engagement rate, and generic content faces up to 50% reach suppression automatically. The mid-2026 twist is that authority and perspective are starting to beat slide decks. This is what a LinkedIn micro content strategy actually needs to earn reach now.
When Stripe first rolled Radar's machine learning out to every merchant, the framing was almost defensive — 'we use ML to stop fraud'. In 2026 that framing has flipped. Stripe has trained a foundation model for payments on tens of billions of transactions, and fraud detection is only one of the things it does. Card testing attacks are down 64%, authorisation rates are up 1.3 percentage points, and Radar now reduces fraud by 38% on average across the network. If you run payments, marketing, or anything that touches conversion, this is the year to pay attention.
When we first wrote about OpenAI's enterprise ambitions, the story was about momentum — early Fortune 500 pilots, a handful of marquee contracts, the sense that ChatGPT was graduating from novelty to infrastructure. Eighteen months later, that framing is obsolete. OpenAI didn't just win enterprise AI — it turned enterprise AI into a category everyone else is now trying to catch up to. 92% of the Fortune 500 are paying ChatGPT customers, 7M+ enterprise seats are deployed, and annualised revenue has hit $25B. The numbers in 2026 aren't projections, they're receipts.
One year ago this month, Microsoft shipped Windows Recall into the wild — the AI feature that photographs your screen every few seconds so you can 'find anything you've ever done' on your Copilot+ PC. The relaunch was supposed to close the chapter with encrypted vaults, Windows Hello and VBS enclaves. Twelve months on, a fresh proof-of-concept exploit has bypassed its most-hyped defence, Microsoft has classified the issue as 'not a vulnerability,' and fewer than 10% of Windows 11 PCs can even run the feature. This is what the first year of always-on AI memory actually looked like — and why marketers building AI workflows should take the lesson seriously.
When we first wrote about GPT-5 last year, it was a leak — a set of rumoured specs and a whispered context window nobody at OpenAI would confirm. A year on, the speculation has resolved into a shipped product. GPT-5.4 launched on 5 March 2026 with a 272K standard context window (1M configurable in Codex), a native Computer Use API scoring 75% on OSWorld, five-level reasoning-effort control and $2.50/Mtok pricing. The interesting story is no longer whether a million-token window is coming — it's what happens to your marketing, your agents and your workflows now that it's the default.
When we first wrote about the AI coding tools 'battle' between Google and GitHub Copilot, the market had two names on everyone's lips. That race is over — and neither presumed finalist won. In 2026, the tool developers reach for first is Claude Code, Anthropic's terminal-native coding agent with 54% market share and a $2.5B annual run-rate. GitHub Copilot still holds 42% of paid seats but only 9% of developers name it 'most loved', versus 46% for Claude Code. Here's the current stack — and why the same pattern is coming for marketing tools next.
LinkedIn finally killed engagement pods. In early 2026 the platform ripped out its entire ranking system and replaced it with 360Brew — a 150-billion-parameter AI model that now detects pods with 97% accuracy. Flagged accounts see reach collapse from roughly 8,500 impressions per post to around 340 overnight, with 60–90 day recovery windows, and pods are now an explicit ToS violation. Here’s what actually works on LinkedIn now — and why the old playbook just became a permanent scar on your account.
When we first wrote about ChatGPT as a mental health revolution, the framing matched the moment. A year and a half later, the numbers tell a different story. In October 2025, OpenAI disclosed that roughly 560,000 users per week show conversational signs consistent with psychosis or mania, and more than 1.2 million discuss suicide with the chatbot. The 2024 revolution framing has given way to a 2026 reckoning — and it reshapes how marketers, platforms and brands must now operate.
On 22 April 2026, Geoffrey Hinton stood on stage at the Digital World Conference in Geneva and told policymakers that the AI industry is "like a car with no steering wheel." It was the bluntest public framing he has offered since he quit Google in 2023. Unlike previous warnings, this one came with numbers, a deadline and a specific industry that goes first. Hinton now puts the probability of an AI takeover at 10–20%, and says AI will have the capabilities to replace "many, many jobs" within about seven months. This piece walks through what he actually said, why his estimates have tightened, and what it means for marketing teams running on systems he is describing.
When we first covered Cloudflare Vectorize v2, the story was about a vector database finally catching up to Pinecone on features. Twelve months later, the story is bigger than the database. Cloudflare has doubled capacity to 10 million vectors per index, launched AutoRAG as a fully managed end-to-end RAG pipeline, and stitched Vectorize into a full-stack AI platform running inside the same edge network as Workers AI inference. If you're building retrieval-augmented generation in 2026, the calculus has changed.
Google I/O 2025 promised an agentic future — Gemini Live, Project Mariner, Project Astra. Twelve months later, Google Cloud Next 2026 (22 April, Las Vegas) shipped the delivery: the Gemini Enterprise Agent Platform, the eighth-generation TPU (TPU 8t for training, TPU 8i for inference), the Agent-to-Agent (A2A) protocol and Workspace Studio. The receipts are in the customer logos — GE Appliances is running 800+ agents, KPMG reports 90% Gemini Enterprise adoption, Tata Steel deployed 300 specialised agents in nine months. This refresh reframes the post as an 'I/O promise vs. Cloud Next delivery' retrospective — and explains why marketing teams now need a vertical operating system on top of Google's horizontal agent stack.
TikTok is no longer a video app with a search bar bolted on — in 2026 it is a search engine with a social feed attached. Google's own research now shows roughly 40% of Gen Z users start product and information searches on TikTok or Instagram, not on Google. TikTok has shipped the infrastructure to match: Search Center inside Ads Manager with a live Search Relevance score, AI Dubbing for instant multi-language shoppable video, audio as a ranking signal, and captions that now beat hashtags by 20–40%. Frame-level alt-text from TikTok now surfaces inside Google's short-video carousel too. If you market a brand, the discovery layer just moved — and TikTok built the map.
When we first wrote about AWS Bedrock, the story was how to call a foundation model from an API. That question is now a footnote. On 22 April 2026, AWS rolled out its biggest Bedrock AgentCore release to date — a Managed Agent Harness, an AgentCore CLI with infrastructure-as-code governance, a Skills system for the major coding assistants, and a Browser that can now click through operating systems. Bedrock isn't an inference API any more; it's Amazon's agent platform. And it declared war on Google and OpenAI on the same day both of them announced theirs.
OneOff, the AI-powered fashion shopping agent we first covered in 2025, has become the most-invested-in platform of the year. On 14 April 2026, Meghan Markle joined as an investor and participant — and her Australia tour looks were indexed and shoppable within hours. She sits alongside Shay Mitchell, Quavo, Aryna Sabalenka, Rob Lowe, Kate Hudson, Emma Roberts, Brooks Nader and the Hermès Family Office in a $4M round that signals a bigger shift: AI is no longer just helping shoppers — it's rewriting how brands reach them.
AI-generated visuals are reshaping how UK businesses explain data. OpenAI’s ChatGPT Images 2.0 sharpens charts and diagrams for professional use. Expect clearer decisions and fewer slide debates.
Amazon’s £25 billion tie-up with Anthropic redraws the United Kingdom cloud map in real time. This deal fast-tracks AI investment into enterprise cloud, and it changes buying power overnight.
Claude Design in the UK lets teams convert simple text prompts into ready-to-use visual assets without hiring a designer. This shift cuts time and unlocks new creative scale — a clear competitive edge.
AI design assistant UK is arriving fast with Anthropic’s Claude Design, a conversational tool for building and refining interfaces. Product teams should pay attention: this changes who does design and how fast it ships.
In the UK, Anthropic is now asking users for government ID and a live photo to access Claude. The policy forces firms to balance data privacy, user verification and AI innovation. Decide whether the price of security is worth the trade.
GPT-5.4-Cyber in the United Kingdom arrives as OpenAI opens limited access to security partners and researchers. The model's permissive design rewrites AI cybersecurity assumptions. Read on to act decisively.
OpenAI in the UK has bought Hiro to fold advanced financial planning into ChatGPT. This purchase forces product and finance teams to redesign advice for scale, fast.
In the UK, MYTHOS model signals a new era for enterprise cyber defence. Security teams must act fast. Now.
Muse Spark in United Kingdom arrives as a signal that Meta is doubling down on Superintelligence Labs and AI model innovation. This launch tightens the race for smarter content, and it matters now.
Muse Spark arrives in the UK as Meta’s latest AI model for personalisation and superintelligence-led features. Product leaders must rethink UX, governance and measurement — the stakes are real.
AI tool terms of service for the UK are suddenly business-critical, not legal fine print. Read the small print and act before a misstep costs trust.
Nvidia acquisition in the UK has put access to Slurm and AI infrastructure under renewed scrutiny. The stakes: uninterrupted training pipelines, researcher access and enterprise roadmaps — decide now.
Generative AI is rewriting how teams design software across the UK. This briefing draws on Anthropic’s Claude and shows how app development can exploit AI growth patterns to launch smarter products. Prepare to think less like an architect and more like a gardener.
Claude AI in the UK has just become costlier for users of third-party tools. Anthropic’s move forces a rethink of subscriptions and integrations — act now.
Automated plant phenotyping is poised to change the United Kingdom’s crop science and farm practices. This fusion of large language models and imaging toolkits could speed breeding and sharpen sustainability—let’s see what’s real.
