{ “@context”: “https://schema.org”, “@type”: “AnalysisNewsArticle”, “mainEntityOfPage”: { “@type”: “WebPage”, “@id”: “https://martech.org/agentic-ai-is-rewriting-martech-Economics-and-infrastructure/” }, “headline”: “Agentic AI rewrites the martech economy and infrastructure”, “description”: “The migration rapid growth from conversational chatbots to autonomous software agents dismantles traditional software licensing models and existing database architectures This analysis evaluates how agentic automation is changing industry cost metrics and forcing marketing operations teams to rethink their core technical stacks. “https://www.linkedin.com/in/pamelaparker/” }, “publisher”: { “@type”: “Organization”, “name”: “MarTech.org”, “logo”: { “@type”: “ImageObject”, “url”: “https://martech.org/wp-content/themes/martech/images/mt-logo.png” } }, “backstory”: “This report is based on pricing analysis software industry technical documentation on multi-agent workflows and enterprise case studies evaluating IT usage patterns and ROI metrics within automated marketing services.”, “speakable”: { “@type”: “SpeakableSpecification”, “cssSelector”: (“h1”, “.article-content p:first of its kind”) } }
Marketers embraced AI when its price was that of an all-you-can-eat buffet. Vendors’ move to token-based pricing comes just as agent workflows have become part of everyday marketing – and agents use many, many tokens. Martech’s infrastructure must change if it is to keep costs low amid growing demand.
As soon as AI connects to your business systems, the chatbot becomes much more powerful. Instead of answering one question at a time, it can pull customer records from your CRM, analyze campaign performance, perform web searches, and generate a custom report in a single workflow. This is made possible by tool invocation, which allows the AI to access external systems via APIs and Model Context Protocol (MCP) connections.
The result is a huge productivity boost for marketers. AI can chain multiple tools together without requiring users to switch between applications. The problem is that each tool call consumes tokens. AI agents, in particular, use an incredible number of them, as they pass the entire task history, their internal reasoning, and all data from external tools through the model at every step of their problem-solving loop.
The reality of the symbolic ceiling
Let’s look at a real-world example of how this works.
A typical daily pipeline (search 200 results, summarize them, generate five stock variations) can easily run 4,000-5,000 tokens or more per run. Over a 30-day month, this can reach well over 100,000 tokens, well beyond the limits of the free tier on OpenAI, Anthropic, and similar platforms, and even enough to pay for a $20 subscription well before the end of the month.
(All token estimates presented in this article are based on standard tokenization metrics used across the industry – the same method vendors use to calculate your bill. These are rough projections, not exact measurements of a live pipeline, and actual usage will vary depending on model, prompt structure, and output length.)
Why Claude Cowork and similar workflows hit the wall
Unfortunately, there is no correlation between the quantity of tokens used and the quality of the result. As Scott Brinker and Frans Riemersma note in the State of Martech Report 2026“more input does not automatically mean a better result” – but you still pay for each element.
Claude Cowork and other tool-intensive environments make the problem visible quickly. Every file read, every search, every API call adds a billable token interaction. Users who start the month with a $20 subscription often find themselves limited by the second week.
The consequence is choosing between limiting your workflow or paying staggering overage fees. Neither is sustainable for a marketing team that must manage pipelines on a daily basis.
The answer is owned context, not a single provider
Fortunately, there is a solution: keep the raw data under your control. Store it in a shared team database like PostgreSQL or Qdrant, in a cloud data warehouse like Snowflake or BigQuery, or in a shared cloud storage folder – and use lightweight, non-LLM filtering logic to extract relevant items before anything hits the model.
This setup might involve a one-time LLM, the same way you might use an AI assistant to write a formula or script. But once set up, it runs automatically on each batch of new data – and doesn’t call an LLM at all. A simple keyword rating or vector similarity search, two orders of magnitude cheaper than an LLM call, ranks data by relevance.
When a social listening pipeline extracts 500 tweets about a brand, the filtering stage discretely selects the 10 most relevant and sends only those to the model. The token bill typically drops by 60% or more. The quality of the information remains the same.
Beyond the one-off agent
There are a number of tools available to perform this type of filtering. THE Open source Hermès agentClaude Cowork, Claude Code, and Perplexity Computer all connect an LLM to external tools, allowing it to call APIs, read files, and automate workflows that might require switching between half a dozen applications. However, Hermes runs on your infrastructure and is vendor agnostic. The others are tied to Anthropic and Perplexity models and infrastructure.
Other notable tools in the broader agent ecosystem include:
- Open Claw(380,000 GitHub stars), a set of open source agents that associate with filesystem-based memory stores;
- OpenAI Codex CLI (93,000 stars), which gives developers access to terminal-based agents with local file persistence; And
- Orchestration frameworks likeLangChain(140,000 stars) andCrewAI(54,000 stars), which you build on rather than using directly.
What they all share, in different ways, is that the model is a guest in your system, not the owner.
Hermès takes this principle to the extreme. It maintains a persistent local context store: your conversation history, tool outputs, and integrations are in your database and accessible from session to session. An additional layer of memory learns from each interaction, capturing preferences, corrections and recurring patterns so that the agent improves over time rather than starting over each session.
Your customers are searching everywhere. Make sure your brand introduces himself.
The SEO toolkit you know, plus the AI visibility data you need.
@media (maximum width: 768 px) { .headline-responsive { font-size: 30px !important; line-height: 1.3 !important; } }
Its integrated ecosystem of tools (web, terminal, API, vision, Python) means the same pipeline that extracts Salesforce or HubSpot records, checks a data warehouse and writes a report, also captures intermediate results and saves them locally. And because it’s vendor agnostic, you only need to change one line of configuration to switch from OpenRouter to a self-hosted LLaMA.
The product is the implementation. The model is what matters – and any team can adopt it. The message is not “use Hermes Agent”. The message is: “start building systems that allow you to own your context, because the vendor-centric approach cannot scale.”
The momentum behind agentic and contextual tools is undeniable. But the real question these tools raise is a strategic one: do you want to pay for the work, or want to own the infrastructure and pay for the reasoning? Upgrade to a larger subscription and you still risk running out of capacity. A different architecture completely removes this problem. The choice every marketing team faces is which side of this equation they want to be on.
This is the first in a three-part series on the transition to agent marketing workflows and the infrastructure needed to support them. In the second part I explain how the architecture works in practice. Part 3 covers getting started with Hermes Desktop: the actual installation, skills and workflows.
The position Agentic AI is rewriting the martech economy and infrastructure appeared first on MarTech.
