Decagon

Decagon started in 2023 with a claim most support vendors only whisper: that an AI agent can actually finish a customer's request, not hand it back to a human at the first hard turn. Founders Jesse Zhang and Ashwin Sreenivas built the whole thing around Agent Operating Procedures, workflows written in plain English so a CX team can change how an agent behaves without filing an engineering ticket.

The market has rewarded that bet fast. By early 2026 Decagon had signed more than 100 enterprises, raised roughly half a billion dollars, and watched its valuation triple to $4.5 billion inside six months. The catch is the one that comes with any two-year-old carrying that price tag: it now has to grow into it.

Decagon sells autonomous AI agents that handle customer support across chat, voice, email, and SMS from a single engine, so a conversation that starts in a chat window can continue on a phone call without losing the thread. The agents don't just answer questions; they take action, pulling data from a CRM, processing a refund, rebooking a reservation, all under guardrails the business controls.

The pitch to buyers is deflection with dignity. Instead of a rigid decision-tree bot, Decagon frames its agents as a concierge every customer gets, and it leans hard on published outcome numbers to make the case.

Decagon's core argument is that legacy support automation is broken in two ways: the bots are dumb, and changing them is slow. Most vendors ship a complex SDK and a black-box implementation, so every tweak to agent behavior becomes an engineering sprint or a support ticket to the vendor.

Decagon's answer is to let non-technical CX teams author and iterate on agent logic in natural language while engineers keep control of integrations, guardrails, and versioning. The company also targets the trust gap: with tools like Watchtower and step-by-step traceability, it tries to answer the question buyers actually ask, which is why did the agent do that.

This is not built for a startup with a shared inbox. Decagon aims at mid-market and enterprise support organizations drowning in ticket volume, the kind that run rigorous RFPs and demand SOC 2, ISO 27001, and hard integration depth before they sign.

It shows up most in industries where support is high-stakes and high-volume: financial services, retail and e-commerce, media and subscriptions, travel and hospitality, and telecom. The buyer is usually a VP or director of customer experience who owns a deflection number and needs to move it without adding headcount.

Decagon's platform splits into three jobs it repeats everywhere: build the agent, optimize it, scale it. Agents ship across voice, chat, and email on one intelligence layer, and a growing set of tools sits on top to keep them honest, with Duet acting as the copilot that turns production failures into fixes.

The newer push is Duet Autopilot, announced in 2026, which promises a self-improving agent that converts live signals into updates automatically, then stages every change for human review before it goes out.

Decagon is classic enterprise SaaS with no public price list. Every deal runs through sales, and pricing is built around usage and business outcomes rather than a per-seat sticker, which is why the company leans so hard on deflection and cost-reduction metrics in its pitch.

That model is also its ceiling. Third-party reviews peg annual contracts well into six figures, which puts Decagon squarely out of reach for smaller teams and keeps its whole go-to-market pointed at the enterprise.

Decagon competes on two fronts at once: legacy support suites like Zendesk, Intercom, and Salesforce that are bolting AI onto older products, and a wave of newer AI-native rivals chasing the same agentic support market.

Its sharpest claimed edges are structural. One engine spans chat, voice, and email instead of a separate tool per channel, and the natural-language AOP approach lets CX teams iterate without engineering, which the company positions as the opposite of the black-box implementations it competes against.

Decagon backs its pitch with hard customer numbers, and it names names. Chime reports 70% chat and voice resolution, one customer saw 10x higher deflection at launch than expected with a 95% cost reduction, and Rippling logged a 32% lift in deflection.

Across the board the company cites figures like 80% deflection, 3x higher CSAT, 65% lower support costs, and 10M-plus customers served. These are the company's own selected wins, so read them as a highlight reel, but the specifics and the named brands behind them are unusually concrete for this category.

The recurring praise is consistent: reviewers say Decagon works better out of the box than chatbots they had tried before, deploys in about a week, integrates cleanly with tools like Zendesk, and comes with a responsive team that ships fixes fast.

The complaints cluster just as tightly. The contracts are expensive enough to rule out smaller companies, real setup still leans on technical agent-building work, and more than one reviewer flags a black-box feeling, where it's hard to see why an agent did what it did. Decagon's own observability tooling is clearly aimed at that last gripe.

Decagon closed a $250 million Series D in early 2026 led by Coatue Management and Index Ventures, tripling its valuation to $4.5 billion in under six months. That came right after a $131 million Series C in mid-2025 at a $1.5 billion mark, so the company roughly tripled its price in half a year.

The cap table reads like an enterprise-AI who's who: a16z, Accel, Bain Capital Ventures, Ribbit Capital, Forerunner, and Elad Gil among them. In March 2026 the company also ran its first employee tender offer at the $4.5 billion valuation, a liquidity move usually reserved for later-stage companies.

Decagon has the rare combination of a real product, named enterprise logos, and a war chest, all before its third birthday. The clearest risks are the ones success created: a $4.5 billion valuation set in a frothy AI market, a crowded field of both incumbents and AI-native challengers, and pricing that locks it out of everyone below the enterprise.

The strategy from here is legible. Push voice and Duet Autopilot toward genuinely autonomous operation, expand internationally, and keep converting deflection metrics into signed contracts. If the outcome numbers hold up at scale, the valuation looks early rather than rich.

Decagon is unapologetically an in-office company, and it wears its values on the wall: Just Get It Done, Invent What Customers Want, Winner's Mindset, and The Polymath Principle. Employees describe an eng-driven, ship-fast environment with real product ownership, the kind of place that suits people who like ambiguity and velocity over process.

The hiring reflects a company sprinting to keep up with its own growth. It's recruiting across engineering, sales, product, design, and operations, concentrated in San Francisco and New York with international expansion into markets like the UK and Australia. Benefits are the standard high-growth package, with a notable addition of fertility and family-building support through Carrot.

Decagon discloses pay ranges on most roles, and they run high. US engineering bands stretch from about $175K to $430K base, with senior individual-contributor and staff roles topping out well past $400K, and product, design, and G&A roles cluster in the $150K to $380K range.

Every role adds equity on top of base, which at a company that just tripled its valuation is a real part of the story rather than a footnote. Bands are also posted in GBP, CAD, EUR, and AUD as Decagon hires internationally.

Decagon sells to financial-services and other security-conscious enterprises, so trust is a product feature, not an afterthought. The company operates as Decagon AI, Inc. and publishes a Trust Center plus a full subprocessor list, and its JDs point to SOC 2, ISO 27001, and GDPR as the compliance bar it engineers toward.

The subprocessor list is a useful tell about the stack: it runs on Google Cloud, AWS, and Azure, routes calls through Twilio, and leans on a wide spread of model providers including OpenAI, Anthropic, Cohere, and xAI. Basis Theory handles tokenized, PCI-restricted data, which signals payment-grade handling for regulated customers.

The headlines through 2025 and 2026 track a company moving fast: a $131M Series C at $1.5B, then a $250M Series D that tripled its valuation to $4.5B, then a first employee tender offer at that same mark.

The product news moved just as quickly, with Duet Autopilot, an experimentation and A/B testing suite, Watchtower, and Agent Versioning all shipping in the same stretch. Founder Jesse Zhang has also been making the public case that in an agentic world, AI agents are never done, reframing the classic build-versus-buy calculus around software that keeps changing after you deploy it.