Typeface on September 22, 2026 made Arc Graph, Arc Fabric and Arc Loop available to its enterprise customers as one connected system, making a company's codified brand guidelines usable inside Anthropic's Claude and OpenAI's ChatGPT and routing campaign results from sources including Google Ads back into the content its AI agents generate next.

In Short

Typeface, which sells AI marketing software to large companies, has switched on three linked tools: one learns how a brand looks and sounds, one runs AI helpers built around the way each team works, and one uses campaign results to decide what gets made next. It matters to big advertisers because their staff increasingly draft material in general chatbots, and a company's style guide can now be applied inside ChatGPT and Claude as well as inside Typeface. What changes is that results from places like Google Ads can flow straight back into the next round of ads and emails, although a person still has to approve anything before it is published.

Three products sold as one system

The Palo Alto company describes the release as the next generation of its Orchestration Engine. According to Typeface, Arc Graph covers brand intelligence, Arc Fabric covers orchestration and Arc Loop covers what it calls closed-loop performance, and the three now operate as a single system that feeds campaign results back into what agents produce. The company says all three are available to enterprise customers from September 22. Neither the press release nor the accompanying blog post gives pricing, packaging, contract terms or regional availability. A live walkthrough of the product is scheduled as a webinar on Wednesday, September 30, 2026, at 10:00am Pacific Time.

Abhay Parasnis, Typeface's founder and chief executive, framed the problem as one of execution rather than insight. "For years, enterprises have had more insight about their customers than ever, but they couldn't act on it. That is the bottleneck we set out to close," he said. "The shift is teaching AI to understand a brand deeply, build agents around how each team actually works, and get smarter as campaigns perform - closing the loop from intelligence to execution and back, so every campaign makes the next one better."

Vishal Sood, founding member and president of research and development, put the diagnosis more bluntly in a company blog post dated the same day. "Performance results die in a dashboard nobody reopens after the read-out," he wrote. Typeface says its platform is used by Coca-Cola, J.P. Morgan, Intel, Chime, Post Consumer Brands and Johnson Controls. The company is backed by Lightspeed, GV, Salesforce Ventures, Madrona, Menlo and M12.

Arc Graph: guidelines turned into structured data

Arc Graph is what Typeface calls its proprietary context layer. The company says it rebuilt the product for this release, moving it from hand-configured rules to a system that extracts, validates and enforces a brand at the moment an AI model generates content. In practice that means converting a set of brand guidelines into structured rules that Typeface's own agents - and now outside AI tools - can apply.

The press release lists five components. A knowledge graph stores what the organization has learned as reusable knowledge that grows with each campaign. Brand ontologies and voices hold several writing voices learned from how the brand actually writes, with what Typeface calls hierarchy resolution for parent brands and sub-brands. An asset management and design system pairs an approved content library, searchable by AI without manual tagging according to the blog, with colors, typography and logos captured as enforceable structure. A Brand Agent checks compliance while content is being created and shows which rule applied and why. The fifth element is the extension of the same intelligence into Claude and ChatGPT.

Most AI tools, Sood argued in the blog, approach a brand like a stranger writing in its voice on day one, and each generic agent a team adds widens the gap between speed and quality.

Arc Fabric: agents configured team by team

Arc Fabric is the orchestration layer. According to Typeface, it spans a runtime that powers agentic work, adaptive agents that carry that work out across a team's channels and systems, and a shared workspace where people and agents create together. Rather than configuring one generic system for an entire marketing department, teams build agents around their own workflows, systems and channels.

Four capabilities are listed. Conversational chat, custom agents and extensibility sit in one place, so a team can run a whole workflow without switching tools; the blog version of that list adds out-of-the-box agents and the collaborative workspace. Agents plan multi-step work, use the team's own systems to finish it and check their own output before handing it over. Repeatable tasks and expertise can be packaged as reusable skills that agents apply again across campaigns. Administrators control data access and how much autonomy agents receive, and the system can be extended through MCP, APIs, webhooks and custom code.

"When one person builds a brilliant AI workflow on their laptop, it solves for one seat," Sood wrote. "The rest of the team is still emailing PDFs."

Typeface claims Arc Fabric brings the power of modern AI coding platforms to marketing for the first time. That priority claim is difficult to sustain. Anthropic extended Claude Code's automation to non-developers through Cowork on January 12, 2026, and in July Anthropic's own marketing operations team described rebuilding its weekly metrics report around that agent, cutting the job from two days to about two hours. There is also a naming gap. In July, Typeface previewed a feature called Chat Cowork, described as a way to hand off a complex multi-step task and return to finished work. The September materials do not use that name, though Arc Fabric's description of agents that plan and complete multi-step jobs covers similar ground.

Arc Loop: results written back into the brand

Arc Loop is the optimization layer, and the product with the most direct bearing on paid media. According to Typeface, it addresses the point at which performance data has long sat in a dashboard while the next campaign is built from best practice. The product pulls results back into Typeface, turns them into what the company calls grounded learnings and feeds those into whatever agents create next.

Three mechanisms carry that out. Results arrive automatically from an ad platform (Google Ads), from marketing platforms (Salesforce Marketing Cloud and HubSpot) and from data warehouses (Google BigQuery and Salesforce Data Cloud), or teams can start with what Typeface describes as a zero-setup CSV upload requiring no engineering. Campaign signals are converted into ranked, explainable content variants, and a person must approve them before anything is published. Each finding is then written back to Arc Graph as reusable knowledge.

Typeface draws the contrast with point tools that optimize delivery inside a single channel. Arc Loop, it says, learns across channels and generates the next campaign from what it finds, "turning performance into new, on-brand content, not just tuned targeting."

On the paid side, though, the connector list is narrow. Google Ads is the only advertising platform named. Meta, Amazon Ads, TikTok, Microsoft Advertising, LinkedIn and programmatic buying platforms such as Display & Video 360 or The Trade Desk do not appear. Results from those channels could in principle reach Arc Loop through a warehouse or a spreadsheet, but the materials describe no such path. Nor do they say how often data refreshes, which metrics are ingested, how learnings are validated before becoming knowledge, or whether ranked variants are tested against a holdout group.

Arc Loop itself was no surprise. When Typeface published its June feature recap on July 8, it previewed Arc Loop as an unreleased feature and described new landing-page analytics in Web Agent as an early step toward it. The idea of routing results back into creative production is not unique to Typeface either. Google's Asset Studio, updated at GML 2026, turns a URL, a brief and goals into a library of tailored static ads built on performance data. Smartly's Creative Insights, added in November 2025, analyzes after a flight why individual ads performed. And HubSpot, one of Arc Loop's named data sources, built its Loop Marketing playbook around a final Evolve stage in which learnings feed the next cycle. What Typeface claims as its distinction is the cross-channel learning and the write-back into a brand knowledge base.

Where the release and the blog diverge

Several details differ between Typeface's press release and Sood's blog post, both dated September 22.

The list of outside assistants is the most material difference. The press release refers to Claude and ChatGPT. The blog names Claude, ChatGPT and Google's Gemini, and adds plugins for Word, PowerPoint and Chrome, reached through the company's MCP connection and other integrations. Component names also shift. The release's Brand Agent appears in the blog as a "brand graph" with an identical description, while "brand ontologies and voices" become "brand systems, attributes, and relationships." The release speaks of three products, yet the product menu captured on the company's blog lists four entries: Arc Graph, Arc Fabric, Arc Agents and Arc Loop. How Arc Agents relates to the other three is not explained.

The framing predates the release. A post on the same blog dated March 4, 2026 already described Typeface evolving into the next generation of its Orchestration Engine and closing marketing's open loop. What changed on September 22 is availability: two months earlier, Arc Loop was still a preview.

An unclosed quotation mark in the release also leaves unclear whether a sentence about companies controlling what AI represents about their brands belongs to Sood; the sentence after it clearly does.

A second 93 percent

According to Typeface's Signal Report research, cited in the release, 93% of marketing leaders say creation, activation and optimization belong in one system, yet fewer than two in five have connected content ecosystems. The release names no edition, sample size or fieldwork dates for either figure.

That omission matters. The Signal Report edition published on June 22, 2026 attached 93% to a different question: the share of leaders saying AI had increased pressure to move faster. That edition surveyed more than 200 marketing leaders at vice-president level or above in May 2026, compared results with a September 2025 fielding, and was described as the fourth in a series. Whether one survey produced two separate 93% results, or the new figure comes from another wave, the release does not say. The fewer-than-two-in-five figure did not appear in the June coverage. Either way, it is vendor research supporting a vendor's product thesis.

The June data frames the release in other ways, some less convenient for it. According to that edition, 86% of leaders said they already used AI agents in campaign execution, yet only 36% had scaled at least one agent, up from 18% in 2025, while 64% remained in pilots. Brand governance was cited as a scaling blocker by 50%, and 37% ranked loss of brand control as their top risk. Only 20% described their workflows as standardized and documented. Arc Graph targets the governance complaint and Arc Fabric the workflow one.

The same research, however, placed the main delay after content exists: 88% said their teams could generate content quickly, with sign-off the obstacle. Arc Loop produces ranked variants that each need human approval before publication. How that approval queue gets shorter is something Typeface has not explained.

Brand standards inside general-purpose assistants

Extending Arc Graph to Claude and ChatGPT addresses a documented gap between governed tools and the assistants staff actually use. Browser telemetry from LayerX found that 47.11% of enterprise AI conversations ran through personal identities rather than corporate accounts, and that users who adopt connectors average 4.5 of them. Copy drafted in those sessions sits outside any brand system unless the guidelines travel with the person writing it.

Typeface is joining an established pattern. Meta opened its ad system to Claude and ChatGPT with write access at the end of April. On August 26, Semrush published a native Claude connector, and the same day Salesforce and Anthropic put 37 prebuilt sales skills inside Claude for pilot customers. On September 14, Snipp Interactive placed brand rebates inside both assistants.

Practical questions remain open. Neither document says which Claude or ChatGPT plans are supported, how Arc Graph authenticates inside them, or whether the assistant enforces the brand's requirements or merely receives them as context. That last distinction has a security dimension. Tool descriptions and tool outputs passed to a model over MCP enter the same context window as instructions, the condition that makes prompt injection possible, and neither the release nor the blog describes controls for that path.

Agents on the other side of the screen

Typeface's forward-looking section argues that brand surfaces will increasingly serve two audiences: people who expect conversational experiences, and AI agents seeking structured, authorized answers. The first stages are in preview with select design partners, according to the company. At one unnamed Fortune 500 brand, Typeface says, marketers build and reshape web experiences conversationally instead of working within rigid templates, compressing work that once took weeks into minutes. No brand name, figures or method accompany that claim.

"Every brand will need experiences built to speak intelligently to agents, not just to people," Sood said in the release.

The release asserts, without citing a source, that agent traffic on the web now rivals human traffic. Measurements from elsewhere suggest a more mixed picture. Cloudflare Radar data for the week ending June 5, 2026 put bots at 57.4% of requests for HTML content, but training crawlers alone accounted for 50.6% of traffic and search-related bots for 10.7%. IAB Australia's crawler guidance, dated July 22, found that only about 2.6% of AI crawler requests were real-time fetches triggered by a person's question. Automated requests have passed human ones. The slice made up of agents acting for a buyer - the audience Typeface describes - is far smaller on those measures.

Customers quoted, without numbers

Two customers supplied statements. "At Chime, everything starts with our members. But reaching millions of people with the right message, at the right moment, while staying unmistakably Chime is an incredibly complex challenge," said Vineet Mehra, chief growth and marketing officer at Chime. He said Typeface gives teams "the ability to move at the speed our members and culture demand," leaving "less time spent on the mechanics of content creation."

Rosa Rita, global chief marketing officer at J.P. Morgan Asset & Wealth Management, described a different goal. "For us, transformation isn't adding AI on top of existing processes - it's rethinking how the work gets done. Our goal is a true segment-of-one experience - where every interaction feels made for that client," she said. Typeface, she added, helps the firm "scale personalized content with brand intelligence and governance to drive measurable business impact."

Neither statement contains a metric. There are no production volumes, cycle times, cost figures or campaign results. For a product whose central claim is that measured performance improves the next campaign, the release offers no measured performance from a named customer. It closes, instead, with a line saying it was created in Typeface.

Why this matters for the marketing community

The release deepens an awkward overlap with suppliers that are also Typeface's partners. Salesforce Marketing Cloud and Salesforce Data Cloud are named Arc Loop sources and Salesforce Ventures is an investor, yet Salesforce made seven prebuilt agents and a long-horizon runtime available on September 11. HubSpot is both a data source and the owner of its own loop framework. Google supplies the only named ad-platform connection, and the relationship is not new: Google Ads partnered with Typeface in 2024 to allow direct asset imports, with templates built for Performance Max, and Campaign Manager 360 added a Typeface partnership in February 2025.

Adobe is the closest comparison on brand codification. In October 2025 it added Firefly Design Intelligence, which codifies design rules into systems applied across Creative Cloud, alongside a Content Production Agent in beta for GenStudio. At Cannes Lions in June, Accenture Song, Omnicom, Stagwell's Code and Theory and WPP committed to deploying Adobe's platforms for enterprise clients. Typeface's own route to service buyers has included partnerships, among them the Agentic Marketing Services offer set up with Firstsource on April 30, the same day it released an AEO Analyzer that scores content for AI answer engines.

Against that cadence - an AEO tool in April, research in June, a feature recap in July - the September release delivers a direction flagged since March rather than setting a new one.

For media buyers the practical scope is narrower than the release's language. Whether Arc Loop's grounded learnings are causal or merely correlational, and how they would interact with the delivery optimization already running inside Google's and Meta's systems, is not addressed in either document. The September 30 webinar is the next scheduled occasion on which Typeface is due to show the product.

Timeline

Summary

Who: Typeface, a Palo Alto-based AI marketing software company backed by Lightspeed, GV, Salesforce Ventures, Madrona, Menlo and M12, with statements from founder and chief executive Abhay Parasnis, president of research and development Vishal Sood, and customers Chime and J.P. Morgan Asset & Wealth Management.

What: Three products - Arc Graph for brand intelligence, Arc Fabric for agent orchestration and Arc Loop for performance feedback - were made available as one system, with Arc Graph extending codified brand guidelines into Claude and ChatGPT (and, per the blog, Gemini) and Arc Loop turning results from Google Ads, Salesforce Marketing Cloud, HubSpot, Google BigQuery, Salesforce Data Cloud or CSV files into ranked content variants that require human approval.

When: The release and blog post are dated September 22, 2026. A product webinar is scheduled for September 30, 2026. Arc Loop had been previewed as unreleased on July 8, 2026, and the orchestration framing dates to a March 4, 2026 blog post.

Where: Availability is stated for enterprise customers with no regional detail; the company is based in Palo Alto, California, and the extension reaches into third-party assistants from Anthropic and OpenAI.

Why: Typeface argues that enterprises hold customer and performance insight they cannot act on, and its own survey work puts brand governance and undocumented workflows among the main barriers to scaling AI agents. For advertisers, the release matters because it ties creative production to campaign results and to the general assistants staff already use, while leaving undisclosed the pricing, the missing Meta and Amazon connections, the method behind its learnings and the source of its headline survey figure.