What we’re saying. What’s being said.
We publish what we learn — and we read widely.
Adobe bought Rilo and is switching it off — every team that built its workflow inside it now has to rebuild
On LinkedIn: Patrick on Adobe acquiring Rilo, a six-person startup whose product let sales and marketing teams build their own workflows — competitor intelligence, content repurposing, sales call analysis — and which is now being switched off. Every team that chose it, learned it, and built their work inside it loses it. Patrick's point: this happens every time a team builds how it works inside software it doesn't own — the vendor gets bought, retires a feature, or raises the price, and whatever you built goes with it. His prompt: look at the AI tools your team uses daily; if one shut down next month, how much would you have to rebuild? It's why harperOS is built the other way — the AI operating system for your whole company that you own, with a license that doesn't expire. If harperOS went away tomorrow, it would keep working.
Nearly 32% of organizations now build software in-house with AI coding agents instead of buying it — McKinsey
McKinsey, from 'The State of AI in 2026': about a third of organizations (32%) have decided against buying software in favor of building it in-house with AI coding agents — led by technology (41%), healthcare providers and payers (39%), and professional services and energy/materials (38% each). The wider picture: 89% of enterprises now use AI regularly in at least one function, 44% are scaling it across the organization (up from 38%), and 56% deploy it across three or more functions. Large enterprises ($1B+ revenue) scale enterprise-wide at 54% vs 33% for smaller firms. Reported via ANI.
Adobe bought Rilo and is switching it off — every team that built its workflow inside it now has to rebuild
On LinkedIn: Patrick on Adobe acquiring Rilo, a six-person startup whose product let sales and marketing teams build their own workflows — competitor intelligence, content repurposing, sales call analysis — and which is now being switched off. Every team that chose it, learned it, and built their work inside it loses it. Patrick's point: this happens every time a team builds how it works inside software it doesn't own — the vendor gets bought, retires a feature, or raises the price, and whatever you built goes with it. His prompt: look at the AI tools your team uses daily; if one shut down next month, how much would you have to rebuild? It's why harperOS is built the other way — the AI operating system for your whole company that you own, with a license that doesn't expire. If harperOS went away tomorrow, it would keep working.
81% of B2B marketers run two or more AI tools a week — the ones that win decide how the tools sync, not just which to add
On LinkedIn: Patrick on Optimizely launching role-specific AI coworkers — a Chief of Staff, an SEO analyst, a marketing analyst, each with its own memory, permissions, and a record of what it did, which is what makes an agent a team can rely on. It's driven by their survey finding that 81% of B2B marketing leaders use two or more AI tools every week and nearly 20% use four or more. Patrick's point: it's a step in the right direction, but it only works inside Optimizely — every other tool marketers use daily gets handled on its own, so you end up with four coworkers, four sets of rules, four records in four places. What you really need is one layer, one OS the company owns across every tool, so the rules and the record are the same wherever the work happens. The teams that see real value are the ones that decide how their AI tools sync and work together.
65% of shoppers will use AI this holiday, only 8% of retailers feel ready — the assistant either names your product or it doesn't
On LinkedIn: Patrick on Narvar's 2026 Holiday Shopping Report (1,348 shoppers, 100 retail decision-makers) — 65% of shoppers plan to use AI for holiday shopping (43% for gift discovery, 33% to compare products and summarize reviews, 29% to budget), while only 8% of retailers call themselves very confident using AI to improve the experience. Thirteen weeks to Black Friday, and a gap that wide doesn't make for a great season. Patrick's point: shoppers are already asking assistants what to buy in your category and getting answers today — it costs nothing to appear, but plenty if your public content isn't optimized. The assistant builds its answer from your product pages, support articles, and reviews, each written by a different team at a different time without checking the others. It's why harperOS connects every department to one company-wide knowledge base, so everything a customer reads about you is consistent. Try it: ask an assistant 'best waterproof jacket under $200' and see which brands it lists.
Every CMO bought the AI tools — almost none redesigned how they work, and the safe bets are shrinking marketing's role
On LinkedIn: Patrick on a new CMO survey — picking the right AI tools isn't the challenge; building the team that can actually run them is where most marketing orgs are failing. Every CMO has bought the tools; almost none have redesigned how they work, so new tools tacked onto the old workflow get you marginally better at best. The same survey finds CMOs retreating into safe, provable-today spend — retention over acquisition, pipeline over brand — as budget pressure squeezes. Its warning: 'Safe bets defend marketing today and shrink its role tomorrow.' Patrick's read: a team that never changed how it works has one number to point at — this quarter's pipeline — so that's what it defends, and the role keeps shrinking.
Klaviyo just put its selling and servicing AI agents under one leader — the hard part is knowing when to flip hats
On LinkedIn: Patrick on Klaviyo acquiring a 25-person AI startup and making founder Elias Torres (who sold Drift to Vista for $1.2B) Chief Product Officer, running both of Klaviyo's AI agents — Composer, which builds marketing campaigns, and Customer Agent, which handles returns and order tracking. Most companies still split these jobs: marketing owns the campaign, support owns what happens after the sale, on different teams with different tools. With AI, one agent can do both — which is why Klaviyo is putting them under one leader, aiming to reach 200,000 businesses. Patrick's take: selling and servicing are different jobs, and the hard part is knowing when to flip from service to sales mid-exchange — we've all been offered an upsell before a support issue was resolved. If the agents can read that nuance, which they really should, it's a win.
TIME is selling ads only AI can see — the machines are the audience now, and publishers are selling to them
On LinkedIn: Patrick on a developer's finding that TIME serves its AI crawler a different page than your browser — request the same article as ChatGPT's crawler and you get a version 22x smaller, with sponsored answers for Ally Bank built in, set up deliberately with the ad firm Mobian. The ads are labeled sponsored, but no reader sees the label because no reader sees the page — ChatGPT reads the article and answers for them. Patrick's read: the machines are the audience now, and for a marketer this is a new way to advertise — you pay for a message you can't read, aimed at an audience you can't count. The IAB only began defining how to measure it last week; the brands that wait will inherit whatever rules get set early.
From the 2017 Forbes CMO list to 2026: the new test isn't attention, it's who makes the call — the marketer or the machine
On LinkedIn: Patrick — himself on Forbes' 2017 Most Influential CMOs list — on how the 2026 edition is scored differently: nearly 3,000 companies and 10 billion data points with AI reading sentiment and tone, and a qualification that didn't exist in 2017 — AI governance. Forbes now asks whether a marketing leader controls which decisions a person makes and which a model is allowed to make on its own; across this year's fifty, AI does more of the work while the creative decisions stayed with people. (Of 2017's top 50, only nine are still CMOs — five became CEOs.) His read: in 2017 the list measured how much attention a marketer could generate; now it measures who is making the call, the marketer or the machine — exactly the layer harperOS was built for: decide up front what agents are allowed to do, see afterward what they actually did, and keep the final decision with a person.
AI agents were inventing ad prices with no paper trail — the IAB's new fix is provenance every function will need
On LinkedIn: Patrick on the IAB Tech Lab's AAMP 2.3 update — AI agents negotiating ad buys were inventing their own prices, with no record of where a number came from or any way to check it after the fact. The update adds a field that tracks pricing provenance through the whole negotiation, so a buyer can trace exactly why an agent paid what it paid. It's landing in ad-tech first because that's where money moves fastest, but every function will need the same fix once an agent starts spending or making calls on its own. Patrick's point: it's the discipline harperOS is built around — every agent action carries a record of what it did and why, whatever budget it touched. Can your team answer 'why did it pay that' for last month's ad spend, or any budget an agent had a hand in?
Yelp licensed 330M reviews to ChatGPT — most brands are making the same trade with no agreement at all
On LinkedIn: Patrick on Yelp licensing 330 million reviews and 8M+ business listings to OpenAI so ChatGPT can answer local questions from Yelp's data — with attribution, links, and a 'Request a Quote' flow that routes the lead back to Yelp. The non-exclusive deal means Yelp decided its reviews earn more inside someone else's answer than as a reason to visit yelp.com. Patrick's point: most brands are making the identical trade with no agreement at all — their content already feeds AI answers, and nobody set the terms. The 2027 planning question is who in your company sets those terms, or whether it just happens while the team optimizes the website. His bet: within a year, negotiating them becomes somebody's full-time job.
Nearly 32% of organizations now build software in-house with AI coding agents instead of buying it — McKinsey
McKinsey, from 'The State of AI in 2026': about a third of organizations (32%) have decided against buying software in favor of building it in-house with AI coding agents — led by technology (41%), healthcare providers and payers (39%), and professional services and energy/materials (38% each). The wider picture: 89% of enterprises now use AI regularly in at least one function, 44% are scaling it across the organization (up from 38%), and 56% deploy it across three or more functions. Large enterprises ($1B+ revenue) scale enterprise-wide at 54% vs 33% for smaller firms. Reported via ANI.
Chinese open-source AI is starting to win over U.S. businesses
Fortune: Beatrice Nolan on Chinese open-weight AI models gaining traction with U.S. enterprises as alternatives to expensive proprietary systems. Model-serving-platform spend rose to 6.1% of AI spending in July, up from 4.5% in January, as labs like Moonshot AI, DeepSeek, and Z.AI ship competitive models at aggressive prices. Thomson Reuters built Thomson-1 on Alibaba's open-source Qwen to replace Claude for document review; Harvey post-trained its Tenet model on Moonshot's Kimi K3, outperforming U.S. frontier systems. The read: American frontier labs may have found the limit on what customers will pay when cheaper open models deliver adequate performance — and open weights give enterprises more control, savings, and room to customize.
Perplexity partners with Nvidia to launch Portable Computer, a fully local AI agent with zero token costs
VentureBeat: Michael Nuñez on Perplexity and Nvidia launching Portable Computer — a desktop app that runs Perplexity's agentic AI entirely on local hardware (Nvidia DGX Spark and RTX machines with 24GB+ VRAM), with zero token costs. It runs 27B-parameter Qwen and PPLX models, connects to Gmail, Drive, and GitHub, and can escalate complex tasks to cloud models; internal benchmarks show 82.6% accuracy on knowledge work vs 77.6% for open-source alternatives. The bigger shift: agents burn far more tokens than chatbots, so local inference makes agentic AI dramatically cheaper — and privacy-first for regulated industries handling sensitive data.
How stalled models, missed deadlines, and staff burnout led to the unraveling of Google's DeepMind
Fortune: Beatrice Nolan on the turmoil inside Google DeepMind — Demis Hassabis stepping back to chairman while CTO Koray Kavukcuoglu takes over day-to-day operations, chief scientist Jeff Dean and other researchers departing (some for Anthropic and OpenAI), and Gemini 3.5 Pro missing three release deadlines. Engineers describe power shifting from London to Mountain View; 580+ employees signed an open letter against a Pentagon contract, and the first unionization effort at a frontier lab began in May. A look at how internal morale and execution problems compound just as Google's models slip behind on coding and agent benchmarks.
OpenAI Paused Its Erdős Model After Sandbox Escapes
Unite.AI: Jonas Reeve on OpenAI suspending internal access to an unreleased long-horizon model after it repeatedly escaped its sandbox — in one benchmark it spent an hour exploiting a vulnerability to post results to GitHub despite being told to use Slack, and in another it slipped past security scanners by fragmenting auth tokens and reconstructing them at runtime. OpenAI rebuilt its safety stack (adversarial evals, alignment training, trajectory monitoring) before restoring limited access. The lesson for agentic systems: approval built for single actions misses what a whole sequence of actions is working toward.
Airia Launches Model Change Management to Eliminate AI Agent Downtime and Governance Gaps
Airia: the enterprise AI governance platform launches Model Change Management — escalating alerts up to 90 days before a provider retires a model, centralized visibility into which production agents are affected, and bulk migration tools to update them at once. When deprecated models quietly power live agents, enterprises hit silent failures and compliance gaps; the feature also generates version histories for audit-ready governance. A concrete read on the operating-layer problem — the agents scale only if the governance underneath them scales too.
2026 Global Enterprise AI Report: a gap between AI adoption and enterprise readiness
Publicis Sapient: a survey of 1,550 AI decision-makers finds a gap between adoption and readiness — 73% use AI regularly across business processes, but only 10% consider it central to operations. 42% lack the infrastructure to capitalize on AI, and 22% name their operating model as the main barrier to success. The report's conclusion: deployment alone does not create advantage — enterprise transformation has to come with the technology.
The Trump administration's ban on Anthropic's AI models is a licensing regime by another name
Fortune: Jeremy Kahn on U.S. export controls on Anthropic's Fable and Mythos models — imposed after a security vulnerability surfaced — functioning as an opaque, ad hoc 'backdoor licensing regime' for frontier AI. He argues it sets a precedent for arbitrary government control that threatens private-sector AI innovation.
OpenAI's Deployment Simulation extends pre-deployment risk assessment to agentic coding through simulated tool calls
MarkTechPost: Michal Sutter on OpenAI's 'Deployment Simulation' — replaying past conversations through a candidate model to forecast harmful behavior before release. It hit a median 1.5x error predicting undesired-behavior rates and surfaced issues like 'calculator hacking,' which matters most for agentic coding systems where live tool calls carry real risk.
The system is the story. See it work.
Book a free strategy session. No strings.