Software Engineer Intern

Zendesk introduces Specialized AI Agents purpose-built for Your Business. September 14, 2026 New Industry and Custom Agents understand your business, and take action across the systems you use to move your business forward. September 14, 2026 - Today, Zendesk introduced Specialized AI Agents, purpose-built for the work that matters most to a specific business or industry. They combine industry expertise with each company's unique knowledge, workflows and connected systems to take action and automate up to 80% of workflows. The launch includes Industry Agents, which come ready for common, high-value work in specific industries, and Custom Agents, which businesses build around their own processes, policies and expertise. Together, they help organizations trust AI with more complex work from start to finish. "The era of generic, one-size-fits-all AI agents is over. The best agents are specialists: deeply grounded in the work, the industry, and the business they serve," said Shashi Upadhyay, President of Product, Engineering and AI, Zendesk. "Specialized Agents give organizations the expertise and context to take on more complex work, earn trust, and drive quantitative impact." Specialized Agents are a key step toward Zendesk's Autonomous Service Workforce, first introduced at Relate 2026. The vision is simple: AI takes on more of the work while businesses remain in control of how that work gets done. Industry Expertise Out of the Box Industry Agents are built for the work businesses within a specific industry have in common. Zendesk is launching with commerce, where AI agents can help customers shop, manage orders, process returns and exchanges, resolve delivery issues, and handle refunds. Over time, Zendesk will expand Industry Agents to additional industries, including financial services, media, and technology, bringing the same industry-specific expertise to the work those businesses handle every day. Industry Agents combine Zendesk AI with the systems, policies, and expertise required to complete high-value workflows from start to finish. In commerce, that means connecting to systems such as Shopify, Narvar, Stripe, and Riskified to understand the customer's request, retrieve the right information, and take action without requiring customers or service teams to move between multiple systems. Zendesk is also working with Riskified to bring industry-leading risk intelligence directly into commerce workflows, helping retailers make smarter decisions about returns, refunds, and exchanges while reducing fraud and policy abuse. "Zendesk Industry Agents are the AI-first platform redefining retail service at scale," said Jeff Otto, Chief Marketing Officer of Riskified. "We're excited to embed our specialist AI agents and Identity Risk Intelligence so that every agent, AI or human, confidently knows who they're actually engaging with, creating delightful experiences for good customers while protecting retailers from fraud and abuse losses." Custom Agents for the Work Only Your Business Knows Not every workflow can come out of the box. Every business has its own policies, approvals and ways of operating. Using Agent Builder - Zendesk's no-code environment for creating, testing and deploying agents - businesses can now build Custom Agents attuned to how their business is run. Customers have already run more than 1 million Custom Agent executions within seven weeks of launch. Some are seeing automated resolution rates spike 10% post-deployment. Businesses define the job an agent owns, the systems it can use, the actions it can take and where human approval is required. Agents can also run proactively in response to business events or on a schedule, extending automation beyond customer-initiated requests. "We want to resolve issues from beginning to end so consumers can get back to using their machines every day," says Leora Machlis, Director of Global Customer Care at SodaStream. "Customizing our integrations helps us act quickly as customer needs change." "When a Toniebox isn't working, it's not just a technical issue - it's a toddler waiting," says Lauren Cooper, Director of Global Customer Happiness at Tonies. "Parents want answers now, not another hold or a long conversation." One Workforce. Any Service Platform. Industry and Custom Agents can run in Zendesk or in service environments including Salesforce and ServiceNow, giving businesses specialized AI without requiring them to replace the systems they already use. As agents complete more work, Zendesk continuously learns from the resulting outcomes, helping organizations understand what worked, where agents needed help and how future service can improve. Featured Research

zendesk verified resolutions. Zendesk now bills on verified resolutions. That is a bigger deal than it sounds. Short answer At Relate 2026 Zendesk launched its Resolution Platform and shifted to outcome-based pricing, billing only for resolutions it can verify rather than ones the AI self-reports. Adding a verification step to the billing path is the most consequential change, because self-graded resolution counts systematically run high. Key takeaways. * Self-reported resolution counts run high. A verification step is the fix. * Zendesk paired it with a Context Graph providing an audit trail for delegated decisions. * Ask your own vendor who verifies a billed resolution, and whether you can audit it. * Verification usually lowers the reported rate, which is a sign of honesty rather than weakness. At Relate 2026, Zendesk launched its Resolution Platform and moved to outcome-based pricing built on verified resolutions. Most coverage focused on the pricing. The more consequential detail is the word verified. The problem verification solves. When an AI system resolves a conversation, something has to decide whether it actually worked. For most of this category's short history, that something has been the AI itself. That is a bad arrangement, and it is bad in a predictable direction. A model asked to grade its own output overstates. When the grade also determines the invoice, the incentive is unambiguous. This is why quoted resolution rates in this market have been close to meaningless. A vendor reporting 70% and a vendor reporting 35% may be describing identical performance, with the difference living entirely in who counted and how. Putting an evaluation step between the answer and the bill breaks that loop. Zendesk pairing it with a Context Graph that captures an audit trail of agentic reasoning and performance context is the same idea applied to explainability: if an agent is going to act on your behalf, you should be able to see why. Why this is good for buyers even if you never buy Zendesk. Because it sets an expectation. Once one major vendor bills only on verified outcomes and publishes an audit trail, "trust our number" becomes a harder position for everyone else. The right response as a buyer is to ask every vendor the same question and compare the answers. The questions: * Who decides a resolution succeeded? The answering model, a separate evaluation, or the customer? * Does abandonment count? A customer who gave up is not a resolution. * Are handoffs billed? Charging for a conversation the AI could not handle means charging for failure. * Does a reopen retract the charge? If the customer comes back about the same issue within a fortnight, the first attempt did not work. * Can I audit a sample? If you are billed on a unit the vendor counts, you should be able to inspect it. A vendor unwilling to answer all five in writing has told you something useful. The number will go down, and that is fine. Expect verification to lower reported resolution rates. Fewer conversations qualify when something checks. That is a feature. A lower honest number is more useful than a higher invented one, because you can plan against it. The failure mode to watch is a vendor whose reported rate does not move when verification is introduced, which usually means the verification is not doing much. Do compare the whole picture rather than the rate. A stricter count at a higher unit price can land in the same place as a loose count at a lower one. The comparison that works is total cost divided by issues genuinely resolved, which Fidiora wrote up in how to build a cost per resolution model. The part that still needs watching. Verification addresses who counts. It does not address what happens when volume spikes. Zendesk introduced automatic overage billing in January 2026, charging teams past their committed monthly volume at the per-resolution rate. Verified or not, a resolution you did not plan for still bills. So the second half of the question is containment: is there a hard spend cap, are you warned before crossing it, and what happens at the ceiling. Verification plus a cap is a model finance can approve. Verification without one is still an unbounded liability with better accounting. Where Fidiora stand. Fidiora has always excluded abandonment, timeouts, and handoffs from billing, and publishes the definition in full at how Fidiora define a resolution rather than describing it in a contract you see after a sales process. There is a hard monthly spend cap you set yourself. Fidiora think Zendesk moving the category toward verified outcomes is straightforwardly good, including for Fidiora, because the whole market becomes easier to compare when everyone has to say who counted. Sources: diginomica on Relate 2026, Futurum Group. Verify current Zendesk rates and definitions on their pricing page. Frequently asked questions. zendesk verified resolutions outcome based pricing resolution platform ai support billing Resolve, don't deflect. See Fidiora resolve a ticket, capture a lead, and keep the bill predictable.

Zendesk adds real-time voice translation for contact centres. Thu, 10th Sep 2026 (Yesterday) Zendesk has introduced Real-Time Voice Translation for Zendesk Contact Centre, bringing two-way translation to live customer service calls. The tool lets customers and agents speak their preferred languages on the same call without switching channels or using a separate interpreter. Voice remains a major part of customer service operations, accounting for 40% of contact centre volume. Language differences can make those calls harder to staff and route. The translation feature is built into Zendesk Contact Centre rather than offered as a separate product. Agents and administrators can manage translation within existing workflows, while businesses use one workforce and one operating system across multiple languages. For contact centres supporting several markets, recruiting multilingual staff can be difficult and costly, especially when demand for a particular language is limited or uneven across regions and time zones. That can lead to longer waiting times, extra transfers and gaps in coverage. The product is intended to reduce reliance on language-specific queues, specialist staffing and interpreters. It also lets teams route calls by issue and agent expertise, rather than sending customers first to a dedicated language queue. Under the setup described by Zendesk, a customer language can be assigned to a call queue or selected by an agent during the call. Agents can switch two-way translation on or off without leaving the conversation, while administrators can decide where translation is enabled and whether to retain the original audio, translated audio or both. Operational pressure The launch comes as contact centre operators look for ways to serve more customers without duplicating teams across languages and geographies. Businesses expanding into new markets often face a trade-off between offering voice support in local languages and maintaining efficient staffing levels when call volumes vary. Zendesk presents the tool as a way to expand language coverage while keeping support within a single operational structure. That could lower the cost of serving multilingual customers and reduce waiting times linked to transfers or interpreter availability. "Customers should not have to choose between speaking the language that feels most natural to them and getting the support they need," said Shashi Upadhyay, President of Product, Engineering and AI at Zendesk. He added that voice support is especially important when an issue is urgent or sensitive. "When an issue is complex, sensitive, or urgent, people want to be understood, not routed somewhere else. Real-Time Voice Translation helps businesses make that kind of voice support available to more customers, while agents stay focused on solving the problem," Upadhyay said. Broader context The addition reflects a wider push across the customer service software market to use artificial intelligence in live service interactions, not only in chat and email but also in voice. While automated translation has been widely used in text-based channels, applying it to live calls raises operational questions around call flow, staffing and record retention. Administrators will be able to control language preferences and decide how translated and original audio are stored. Those choices may matter for organisations balancing customer access with internal quality control and record-keeping requirements. The product will first be offered to eligible users of Zendesk Contact Centre Native through a closed early access programme. The initial launch suggests Zendesk is testing adoption within its existing contact centre customer base before any broader rollout. The move also signals the importance Zendesk places on voice within its customer service portfolio at a time when many software providers have focused heavily on AI agents and self-service tools. By targeting live multilingual calls, it is addressing one of the more labour-intensive parts of support operations. In practical terms, the feature is designed to let a single pool of agents handle a wider range of customer conversations rather than requiring dedicated teams for each language. For companies with fluctuating demand across markets, that model could offer greater flexibility in scheduling and routing. Zendesk said one workforce could serve more customers in more languages through a single system and set of workflows.