In May 2026, 130 agency owners flew to Austin for Vistara, E2M’s annual gathering on the future of agency work. Most AI events draw vendors, analysts, and the technically curious. This one drew operators: founders running fifty-person shops, husband-and-wife teams scaling boutique outfits, and executives from agencies you’ve heard of and many you haven’t.
Two days of sessions made one thing clear. The conversation has moved past whether AI matters for agencies. Jason Swenk put it plainly from the stage: “AI isn’t an advantage anymore. It’s a requirement.”
What follows are the five signals we took from Vistara 2026, drawn from the speakers who led each topic. Together, they describe a structural shift in how agencies are built, sold, delivered, and discovered. We’ve kept this overview tight; each signal opens into a deeper piece later in the series.
Signal 1: The Agency Model Is Breaking
Brent Weaver opened the conference with an exercise. He asked the room to write down their biggest dream and the destination they were building toward. Then he held up the agency models most people were using to get there. Some, he said, were trying to reach the moon on a scooter.
The metaphor stuck. Every speaker after him reinforced it. The agency model that worked through 2024 (hourly billing, project-based delivery, you the founder as the quality bottleneck) is structurally mismatched to what AI now makes possible. E2M co-founder Manish offered the cleanest analogy: ATMs didn’t kill banks. They freed tellers to sell wealth management and mortgages. The agency equivalent is freeing your team from delivery to sell strategy and outcomes your clients didn’t know they needed.
Jason Swenk mapped the identity shift required. Operator does everything. Manager hires people but still does everyone’s job. Architect builds systems that decide without you. CEO sets vision. Owner is accessible but optional. Most founders are stuck at Operator or Manager. AI is what lets you skip levels.
Underneath the strategic shift sits a quieter pressure. The build moat is eroding. Britney Muller spent her session showing non-technical marketers shipping production tools in a week. Jennifer McPherson, who has “never written a line of code in my life,” is building a SaaS platform for the AEC industry. Dale Bertrand described losing a $7,500-per-month deal to a competitor whose account manager was, literally, a chat agent. The infrastructure assumptions of human-paced work do not survive agent-paced workloads.
This is what’s coming for every agency that doesn’t redesign first.
Signal 2: Outcome-based Pricing Is Forming A New Commercial Layer
If the model is breaking, the question becomes how to sell what replaces it. Dale Bertrand of Fire & Spark spent his Vistara session on exactly this question. His framework is the most cited single contribution of the conference.
Start with training as discovery. Charge for it (paid out of the client’s training budget, which sidesteps procurement) or give it away. Either way, you’re learning what to pitch while the client is learning to think differently about AI. Move into a 30-day pilot with outcome-based pricing. The deal: hit the productivity target and you get paid. Miss it and you don’t. Bertrand only offers this on workflows he’s already proven internally, so the risk is calculated. The result is a client who’s emotionally invested in the outcome and procedurally locked into a longer retainer.
His worked example: a global bank engagement that started as a single internal-linking agent and expanded to $140,000 over 14 months. They replicated the agent across 12 regions in 12 languages in the first week. None of that required new building. It required the right pilot.
McKinsey announced this year that it’s moving to outcome-based pricing across consulting. Bertrand’s framing predicts where the rest of the market goes next. Adam Burrage at Trident demonstrated a smaller version of the same arc: an automated front-end sales-to-proposal workflow that he packaged and sold as a $5,000 pilot, scaling to $28,000. The money, in Burrage’s phrasing, is in the connection and the automation, not in writing prettier emails.
There’s an undercurrent here worth naming. As vibe coding compresses the cost of building, hourly pricing collapses. You either race competitors to zero or you sell something hourly pricing cannot measure: outcome, certainty, accountability. The agencies winning this transition are the ones charging for the dinner, not the recipe.
Signal 3: Production AI Delivery Is Real, Not Theoretical
The most underestimated takeaway from Vistara is that the workflows people are debating online are already running in production at the agencies represented in the room. Ronik Patel, who leads AI adoption at E2M, walked through three of theirs.
The first: a content workflow that takes a single keyword through DataForSEO and SERP scraping into an LLM-generated brief. It saves the SEO team roughly 600 hours per month. The second: a Figma-to-WordPress pipeline using Claude Code, custom skill files, and Playwright for automated QA. Each page run fires dozens of parallel requests through the model, the browser, and the underlying WordPress instance. The throughput profile bears no resemblance to a human developer’s. The third: a self-serve onboarding workflow that lets agency clients pick an Elementor template and receive an AI-generated preview before development begins.
Khushbu Doshi, E2M’s COO, introduced what she calls the Agency AI Intelligence Layer. The core insight: the same agent, given the same prompt, produces dramatically different output when paired with a per-client context file capturing history, what’s worked, what hasn’t, and what’s been refused. Memory is the difference between an agent and a calculator.
Bryan Sekine and his wife at Fox & Allen Marketing built seven AI tools to support their real-estate SEO niche, then partnered with E2M to unify them into one dashboard. Their onboarding went from four weeks to three to a target of three days. Agency-wide AI adoption climbed from 25 percent to 85 percent. Jennifer McPherson, also a non-developer, is building RFP Pursue AI as a full SaaS platform for the AEC industry.
These are not pilots or demos. They are running, billing, and replacing manual work today.
Signal 4: Discovery Is Shaped by AI
Andy Crestodina, co-founder of Orbit Media, used his session to push agencies past a comfortable misframing. The conversation, he argued, is not about AI visibility. It’s about AI recommendations.
The distinction matters. Search delivered users a list. Users did their own consideration thinking, hit the back button, and converted on the next page. AI delivers users a recommendation. The consideration thinking now happens inside the model, in conversation, before the click. By the time a visitor arrives on your site from ChatGPT, they’re already pre-sold; the website’s job has shifted from generating traffic to closing a conversation that started elsewhere.
This collapses two disciplines that used to be staffed by different people. SEO optimized for systems. Conversion copywriting optimized for humans. AI optimization requires both at once. Crestodina demonstrated a synthetic-prompt audit that predicts how a buyer would prompt for your category, then identifies the gaps in your content that prevent the model from recommending you.
His most quotable correction: stop using SEMrush to find Frequently Asked Questions. Use your sales call transcripts. The voice of your actual customers is sitting in every call recording you’ve made since 2020. SEMrush knows what strangers ask. Your transcripts know what buyers ask.
For agencies, this means rebuilding service pages, case studies, and home pages to address buying criteria explicitly. AI doesn’t infer your strengths. It quotes what you publish.
Signal 5: The Infrastructure Tax
The signal almost nobody on stage named directly was the one running under every demo: agentic AI changes what you need from the stack underneath it.
Token economics came up repeatedly. Ronik Patel mentioned E2M spending around $250 per day on tokens for production workflows. Dale Bertrand’s healthcare content pipeline costs roughly $35 per article in inference. Most speakers were absorbing these costs to compete on price, but acknowledged the math gets harder as workflows scale.
The quieter cost is throughput. An agent-paced workflow does not behave like a human-paced one. When Ronik’s Figma-to-WordPress pipeline runs, it fires concurrent requests through Claude Code, Playwright, the WordPress admin interface, and the page build process simultaneously. Bryan Sekine’s seven-tool dashboard runs similar loads. Automated content generation at the volumes Ronik described, 500 to 2,000 articles per month, does the same. These workloads are concurrent, persistent, and unforgiving of infrastructure designed for a human developer hitting save and waiting.
Most hosts cap PHP workers. You don’t see the ceiling until your first production agent hits it, usually under deadline pressure with a client watching.
At Servebolt we don’t cap PHP workers. As workloads shift from human-paced to agent-paced, that’s the kind of constraint that stops being a footnote and starts becoming the bottleneck between your agentic ambitions and what your stack can actually deliver. It’s a topic we expect to keep returning to as this series, and our own product roadmap, unfolds.
The Agencies That Win Have the Right Infrastructure
Five signals, one underlying shift. The agency that wins 2026 is not the agency that builds fastest, charges least, or adopts the most tools. Those races are already decided in favor of whoever runs a scrappier setup with cheaper labor.
The agency that wins is the one that combines three things at once. A model designed for agent-paced delivery, not retrofitted to it. A commercial layer that sells outcomes rather than hours, because hours stopped being measurable as soon as Claude Code could compress them. A discoverability stance that assumes the buyer is being recommended to, not searching for, your services.
What sits underneath all three is rarely discussed and increasingly important: infrastructure designed for the workloads agents actually run. We’re watching this layer carefully because it’s where Servebolt has been positioned for years without naming the use case. We don’t think of ourselves as an agentic host. But the agencies running production workloads on our infrastructure are quietly discovering that the constraints they’d accepted elsewhere don’t exist here.
The rocket ship Brent Weaver described in his opening is being built. We’re not selling agencies the ship. We’re making sure the launch pad holds.
Where to Learn More About AI for Agencies
Each signal above opens into a deeper piece. Over the coming weeks we’ll talk about:
- The new agency identity, vibe coding pressure, and what comes after the holding-company model
- Outcome-based pricing in practice, the Race to Zero, and how to charge for AI services clients don’t yet understand
- Real production AI workflows, with the tools, prompts, and skill files behind them
- The full GEO and AEO playbook for agencies
- The agentic infrastructure stack and where it’s going
Huge thanks to the Vistara 2026 speakers whose work informed this overview: Brent Weaver, Britney Muller, Dale Bertrand, Emma Jackson, Ronik Patel, Khushbu Doshi, Jason Swenk, Andy Crestodina, Jennifer McPherson, Adam Burrage, Bryan Sekine, Vishal Mahida, and David C. Baker. Special thanks to E2M for convening the conversation.
