Not evolution — metamorphosis. Marketing technology is dissolving and reassembling like a caterpillar in a chrysalis. Scott Brinker & Frans Riemersma map the five dimensions of structural transformation from campaign managers to value engineers, from SEO to AEO, from SaaS to Context-as-a-Service.
The old forms: deterministic SaaS, campaign managers, system administrators, marketer-controlled channels. Comfortable, familiar, but not the final form — just a caterpillar with better legs.
Where most organizations are now: the messy middle. Old structures dissolving, new ones assembling. AI everywhere but integrated nowhere. Stack wranglers and multimodal operators navigating the soup.
Where this is heading: Context-as-a-Service, value engineers, context engineers, orchestrated intelligence. The customer's AI agent negotiates with the brand's AI. Same DNA, unrecognizable form.
| # | Caterpillar (Past) | Chrysalis (Present) | Butterfly (Future) |
|---|---|---|---|
| 1 | Marketer controls — owned channels, managed funnels, SEO as an instrumentable game | Customer intermediates — AI search mediates discovery, conversations in ChatGPT/Claude/Gemini | Customer's AI agent — researches, compares, negotiates, purchases on buyer's behalf |
| 2 | Isolated AI tasks — generate this email, summarize this meeting, create this image | AI everywhere, integrated nowhere — every SaaS has AI, early agentic workflows, one task at a time | Orchestrated intelligence — agents work across systems, maintain context, act autonomously within guardrails |
| 3 | Deterministic SaaS — subscription platforms, rule-based workflows, fixed machine | Great rebundling struggle — incumbents bolt AI on, AI-natives compete on models, buyers hedge | Context-as-a-Service — platforms deliver right data/content/capabilities to right agent at right moment |
| 4 | Campaign manager — plan, execute, measure, repeat | Multimodal operator — juggling AI tools, legacy workflows, expanding surface area | Value engineer — designing systems delivering measurable value across every customer interaction |
| 5 | System administrator — configure platforms, maintain integrations, manage permissions | Stack wrangler — stitching SaaS + AI + custom, duct tape and good intentions | Context engineer — orchestrating what reaches each AI agent at the right decision moment |
Only ~30 Shopify merchants onboarded. Walmart's EVP called it "a very temporary moment in time." OpenAI pivoting to app-based commerce where merchants own checkout.
Downloads collapsed from 3.3M to 1.1M. Burning ~$1M/day in compute. The $1B Disney licensing deal dissolved. Six months from world-changing to shuttered.
App Directory launched, third-party submissions open, 800M weekly active users. The largest app cluster: martech vendors. Early innings but structurally significant.
"The surface area where AI intersects marketing keeps expanding. It's the individual plot points that are unpredictable, not the arc."
The Model Context Protocol is emerging as the standard for agent-to-tool and agent-to-agent communication. The largest cluster of ChatGPT Apps are martech vendors — Adobe, Airtable, Amplitude, Asana, Canva, Clay, ClickUp, Common Room, Conductor, Figma, Gamma, HeyGen, HubSpot, Intuit Mailchimp, Jotform, Klaviyo, Lovable, Notion, Replit, Semrush, Slack, and ZoomInfo. This creates a composable AI stack where specialized tools plug into a common agent orchestration layer rather than locking into monolithic platforms.
Zero-sum framing (AI replaces marketers) vs. abundance framing (AI amplifies what marketers create). The report argues for abundance backed by evidence.
Internal tools, workflows, data vs. what customers experience. AI forces alignment — agent-mediated discovery makes internal inefficiencies externally visible.
Adding sound to film transformed the entire medium, killed careers, created new ones. AI in marketing is the same order of transformation, not a feature addition.
The martech stack splits into layers (infrastructure, data, intelligence, experience). B2B leads on breadth; B2C builds for depth. Build vs. buy is the wrong question.
AI adoption follows a trust gradient — low-trust tasks (content generation) first; high-trust tasks (autonomous budget allocation) face steeper adoption curves.
AI capability has outpaced organizational policy. The gap between what AI can do and what is responsibly permitted must be closed with guardrails, audit, and escalation.
Retrieval-Augmented Generation is the hidden connective tissue — injecting brand context, customer data, and governance into AI interactions across every channel and tool.
Search optimization reinvents for AI answer engines. The goal shifts from ranking for clicks to being cited accurately in ChatGPT, Claude, and Gemini responses.
| Old Role | Transitional Role | Future Role | Description |
|---|---|---|---|
| Caterpillar Campaign Manager | Chrysalis Multimodal Operator | Butterfly Value Engineer | Designing systems that deliver measurable value across every customer interaction |
| Caterpillar System Administrator | Chrysalis Stack Wrangler | Butterfly Context Engineer | Orchestrating what data, content, tools, and instructions reach each AI agent at decision time |
| Caterpillar SEO Specialist | Chrysalis AI Visibility Manager | Butterfly Agent Presence Architect | Ensuring brand accuracy and citation quality across AI answer engines and agent-mediated discovery |
"Forward-looking marketers are increasingly becoming managers of AI agents and agentic workflows rather than campaign creators."
— Raviteja Dodda
"Humans are still ultimately the ones consuming the message. You need to build for agents, but understand that humans are still the end consumer."
— Sara Faatz
"Underneath all the AI noise, I think we're going to see a renaissance in how we really talk to our audience — the best way, the most personalized way, the most human way."
— Brendan Farnand
"Data silos are the enemy of AI. Your AI is only as good as your data. It doesn't matter how much money you spend on a wonderful AI tool if your data is not connected."
— Tara DeZao
"In contrast to coding with AI, marketing is not a verifiable domain. You can't send 100,000 emails to someone until they purchase a product."
— Anthony Rotio
"The shift from traditional CDP capabilities to CDPs becoming context-ready decision layers is natural and what the market is demanding."
— Jonathan Moran
"Marketing is fundamentally a role of taste, judgment, and creativity. Marketing teams will be among the least affected by AI headcount reductions."
— Tejas Manohar
Stop framing change as gradual improvement. Audit your stack, roles, and customer touchpoints for caterpillar/chrysalis/butterfly state. Plan for discontinuous change.
Treat context as a first-class asset. Map what data, content, and instructions each AI agent needs at decision time. Build metadata layers, knowledge bases, and RAG pipelines.
Your next customer may arrive via their AI agent. Implement AEO/GEO — structure your brand and content so AI answer engines cite you accurately.
Replace "build vs. buy" with "where do we need reliability vs. flexibility." Each stack layer — infrastructure, data, intelligence, experience — needs different investment logic.
Build governance frameworks for agent autonomy — guardrails, audit trails, escalation paths. The speed of AI capability is outstripping organizational readiness.
Start with low-trust AI tasks, graduate to higher-trust applications. Evolve marketers into value engineers. Marketing requires taste, judgment, and creativity.
Evaluate vendors on MCP support. Build agent workflows that compose specialized tools through a common orchestration layer rather than locking into monolithic AI platforms.
The chrysalis — a structure a caterpillar builds before dissolving and reassembling as a butterfly. Marketing technology is not evolving; it is undergoing structural dissolution and reassembly. Same DNA, unrecognizable form.
1) Who controls the conversation (marketer → customer's AI agent). 2) AI in marketing (isolated tasks → orchestrated intelligence). 3) Martech software (SaaS → Context-as-a-Service). 4) Marketing roles (campaign manager → value engineer). 5) Marketing ops roles (system admin → context engineer).
The butterfly-state of marketing ops — orchestrating what data, content, tools, and instructions reach each AI agent at the right moment. Context engineers make agentic marketing actually work by ensuring AI agents have the right context at decision time.
The Model Context Protocol is emerging as the standard for agent-to-tool communication. The largest cluster of ChatGPT Apps are martech vendors — Adobe, HubSpot, Canva, Notion, Clay — creating composable AI stacks where specialized tools plug into a common orchestration layer.
Agent/Generative Engine Optimization — optimizing brand presence for AI answer engines (ChatGPT, Claude, Gemini) rather than search result pages. The goal shifts from ranking for clicks to being cited accurately in AI-generated responses.
Neither — it is stratifying into distinct layers (infrastructure, data, intelligence, experience) with different dynamics. B2B companies adopt broader stacks; B2C builds deeper specialized stacks. Build vs. buy is the wrong question.
Instant Checkout was sidelined after ~30 merchants onboarded. Walmart's EVP called it "a very temporary moment." Sora shut down after downloads collapsed (3.3M → 1.1M) while burning ~$1M/day in compute. Both illustrate AI's unpredictable fast lane.
AI adoption follows a trust gradient: low-trust tasks (content generation, summarization) are adopted first. High-trust tasks (autonomous budget allocation, customer-facing decisions) face steeper adoption curves because the consequences of error are greater.
Retrieval-Augmented Generation is the hidden connective tissue — injecting brand context, customer data, content, and governance rules into AI interactions across every channel without training on proprietary data. RAG makes AI useful by giving it access to context.
Anthony Rotio's observation: unlike coding (where tests verify correctness), marketing outcomes can't be verified by running a test suite. You can't send 100,000 emails until someone purchases. Marketing AI requires judgment, not just verification.
Scott Brinker (HubSpot VP, "godfather of martech," author of Hacking Marketing) and Frans Riemersma (MartechTribe founder, 30+ years consulting). Sponsored by GrowthLoop, Hightouch, Knak, MoEngage, Pega, Progress, and SAS. Design by Angela Ribeiro da Silva.
The shift to buyer-centric marketing is no longer optional. When the customer's AI agent evaluates your product and decides whether to surface your brand, the buyer's context is the playing field. Prepare for agent-mediated discovery, context engineering, and governance frameworks.
The structural dissolution and reassembly of marketing technology — not gradual evolution but chrysalis-level transformation.
Orchestrating what data, content, tools, and instructions reach each AI agent at the right decision moment.
The martech ecosystem standardizing on Model Context Protocol for agent-to-tool and agent-to-agent communication.
Agent/Generative Engine Optimization — optimizing brand presence for AI answer engine citations rather than search rankings.
The vendor ecosystem appearing "flat" in aggregate but churning beneath — the stack is stratifying, not consolidating.
Marketers as managers of autonomous AI agents and agentic workflows rather than campaign creators.
The gradient of AI adoption from low-trust tasks (content generation) to high-trust tasks (autonomous budget allocation).
Architecture injecting proprietary context into AI interactions without training on it — the hidden connective tissue of martech.
The future marketing role — designing systems that deliver measurable value across every customer interaction.
The growing disconnect between AI capability and organizational policy, controls, and accountability structures.
Massive value creation and destruction happening simultaneously as old martech models dissolve and new ones assemble.
The stack splitting into distinct layers rather than consolidating — different dynamics, vendor sets, and buyer behaviors per layer.