Optimizely

🗽 🍎 New York, September 1st. Stella, Senna and I are taking over a room at the Jarvis Center to teach marketers how to build AI agents. Join us! I've been to a lot of AI keynotes this year and the pattern never changes. Someone runs a demo. The room claps. Everyone photographs a slide and flies home. The slide is still in your camera roll. The workflow is still manual. Opticon is skipping that part because our Agent Lodge is a workshop room. Nothing to photograph, because the thing you take home is the thing you built. I've taught 3000+ marketers through Opal University this year and day one always goes the same way. Everyone can define an agent beautifully. Ask who has shipped one and the room becomes deeply interested in the carpet. Alexander has been circling this all year, and his read on 2026 is that AI handed every team the same models and the same speed, so the only edge left is whether anyone can tell the work came from you. (He's the one in the photo, by the way.😉) On the 1st he gets to introduce THE Seth Godin at Opticon!!! 💜🐮 iykyk So pick a workflow you're bored of and come find me for a coffee at Optimizely's Opticon! Link to register in the comments. (it's free!)

Your AI shouldn't just respond to prompts. It should run your workflows. The latest Optimizely Opal release puts you in the driver's seat, building, running, and governing AI agents natively inside Opal Chat. Inside this release: — Agent Builder & Skill Builder: configure and refine agents directly inside Opal Chat, no dev team required — Google Ads Account Auditor Agent: pulls campaign data, flags waste, identifies scaling opportunities, delivers prioritized actions — Four new connectors: Google Search Console, Gmail, HubSpot, ZoomInfo, plus support for remote MCPs — Governance built in: safe URL browsing, version history, execution status visibility, agent update warnings — Nested workflows & static/predicted parameters: more control over how agents chain and execute This is AI as infrastructure, not just a co-pilot. Dive in: https://lnkd.in/eMb6QdtN

The agenda just dropped. Camp Opticon. NYC, September 1. Whether you're deep in the weeds with AI or just figuring out what it can actually do for you, this is where digital and experience makers come ready to build. Live builds around the campfire. Create and optimize digital experiences, live — agentic tools building real work in front of you. Hear from the people already out on the trail, testing and shipping with it. See the agenda and register here: https://lnkd.in/dxVyjzF7

“The AI works in the background. The UX keeps the user in the foreground.” That’s the line from Amit Shimshi’s post that I couldn’t agree with more. It sounds simple, but the gap between those two things is where enterprise AI actually earns trust or loses it. When something goes wrong in a real workflow, a flawed brief, a missed segment, a stalled handoff, users need something to hold onto. Not a full audit trail. Just enough to understand what happened and correct course. That isn’t a model problem. It’s a design problem. What does that look like in practice? Agents that show which skills they loaded and what they searched before the answer lands. Structured inputs that close the gap between “what does this agent need?” and “what am I supposed to type here?” A library organized around actual tasks instead of capabilities buried behind admin gates. None of this breaks the ambient experience. The machinery can disappear. The feedback loop cannot. Amit calls it The Invisible AI Trap 👇👇👇 https://lnkd.in/g6Cg9Rkz It's so good, so I can't wait until #OpalTuesday Optimizely AI #AIUX #EnterpriseAI

Your site has two audiences now.
 Buyers want an experience. AI agents want an answer they can cite. The balancing act between human experience and machine readability is the new strategic question.
 Join us for Content + Beyond.
 📅 20 Aug | 10 AM–12 PM EST   What you'll learn:
 — Create for humans: how to make every visit worth the trip — Create for agents: real AEO lessons from marketing leaders — Keep your cool: the honest playbook for an AI-built website from our own marketing team   Save your seat: https://lnkd.in/eTjCYXyp

"This wasn't a little refresh, this was a complete reinvention." Have you met the new Optimizely? The AI platform for marketing. Optimizely's Gregory Heinz and Dan Reid joined Andi Robinson, host of Behind the Campaign, to talk about the rebrand — and the origin story of Mark Eting. Listen in here: https://lnkd.in/ewpW9peE

Sydney, clear the diary Hear how trailblazers are harnessing AI and experimentation to build intelligent, high-impact digital experiences. Live at Optimizely Digital Summit Sydney. This year's theme: The AI in Brainpower — AI as a force multiplier for human creativity. A full day of keynotes, customer spotlights, hands-on breakouts, and honest conversations on scaling the doing to free up the thinking. When? Thursday, August 13 | ILUMINA, Sydney Register now: https://lnkd.in/eXH2yU8x

Managing autonomous agents is a lot like managing human teams. You don't fire a creative marketer the second they take a wrong step on a project, you give them a course correction. So why do most agent reliability strategies stop at pass/fail? Evals should be the floor, not the ceiling. Here's how we're building a holistic AI reliability stack at Optimizely! TLDR; 🎭 Output quality and execution quality are different things. An agent can produce good outputs through bad patterns. Those patterns are usually the first sign something deeper has shifted. 🧱 Agent reliability requires multiple layers, not one. Evals, input/output guardrails, and behavioral monitoring are complementary. Each catches what the others miss. ⏱️ Behavioral guardrails monitor how an agent works, not just what it produces. By watching execution patterns in real time, you can detect problems as they unfold and rescue runs before they fail. 💸 Runaway agents are a real cost problem. At enterprise scale, an agent stuck in a loop doesn't just produce bad output, it burns through tokens and API calls until something stops it. Catching bad patterns early is as much a financial decision as a quality one. 🌱 Learn before you enforce. Anomaly detection systems should build baselines from successful executions, not guess from day one. That's the core principle of the Learn-Observe-Enforce-Rescue framework. 🛟 Most systems know how to stop a failing agent. The best ones know how to save it. Detection without recovery is just a more expensive failure. The bar for autonomous agent reliability isn't blocking bad runs, it's restoring order when they go wrong. Full article in the first comment 👇

Sydney, I'm coming back - and I'm counting down the days! I can't wait to be back Down Under on August 13th for our annual Digital Summit. The trips to Australia are always special as we have such a fantastic customer base, vibrant partner community, and an incredibly dedicated team. I would love to see all my Australian friends and make new ones - so join us on the 13th or ping me to find time to meet 1:1. Link to Summit is in the comments. Tracy Chalk Viviane Girgis Jonathan Goh Amit Shimshi Michiel Dorjee Simon McDonald Annie Wells Deepali Saini Joseph Tesvic Marty Drill Paul Dearlove Anthony McMahon Christopher Koutis John Ruthven Jason Phillips Dylan Alcott AO Bettina McMahon Mark Ryan Betsie C.