Writer launches an’super agent,’ which actually gets the job done and outperforms OpenAI in key benchmarks

How Writer’s super agent performs tasks that other AI can only describe

Action Agent is fundamentally different from other AI assistants because it operates at “level four orchestration”the highest level of AI automation. Most enterprise AI tools are at level one or two and can only handle basic tasks such as answering questions or retrieving files. In an interview with VentureBeat, Matan-Paul Shetrit explained that most AI tools are at levels one or two. “What we have done here is full orchestration.” This is an agent who can call agents, write its own tools, and execute that with full transparency.”

It goes beyond simple automation. While traditional AI assistants such as ChatGPT Shetrit noted that Copilot is “very much built to like a Q&A experience.” Action Agent was designed for execution. “The difference is that one doesn’t just do back and forth brainstorming but rather, if I do the brainstorming I can also take action.”

Each session, the agent operates in its own isolated virtual machine, allowing it browse web pages, create software, solve technical issues, and execute complex multistep plans. When asked to perform, for example a product analysis. Action Agent automatically processes thousands of customer reviews. It performs sentiment analysis, identifies themes, and creates a presentation – all without human assistance.

When existing tools are not sufficient, the system can generate its own. Shetrit explained that the system can act whether or not MCP or other tools are available. It can generate its own tools for the task on the fly.

Shetrit demonstrated the agent selecting a clinical trial site, a process which typically takes weeks of human research. The agent systematically analysed demographics in multiple cities, ranked them by suitability criteria and generated comprehensive reports. Shetrit noted that the companies had put in weeks of work. “It is not trivial work.”

Breaking benchmarks: action agent outperforms OpenAI in key tests

Writers’ claims about the agent are backed up by impressive benchmark results. Action Agent scored 61% in GAIA Level 3 is the most challenging benchmark of AI agent performance. It outperforms competing systems such as OpenAI’s Deep Research The agent scored 10.4% on the CUB leaderboard (Computer Use Benchmark), making it the best performer in computer and browser tasks.

The results show the agent’s ability in handling complex reasoning tasks, which have traditionally stumped AI-based systems. The GAIA Level 3test requires agents to navigate multiple tools and synthesize data from different sources. They also complete multi-step work flows. This is exactly the type of work enterprises need automated.

This performance is derived from Writer’s Palmyra Modelhas a context window with a million tokens, allowing it to process hundreds pages of documents at once while maintaining coherence in complex tasks. This massive context capability enables the agent to work on entire codebases and research reports as well as comprehensive datasets while maintaining coherence across complex tasks.

The enterprise focus of Writer sets it apart from other AI companies that are trying to adapt their products to business use. The company built Action Agent is now available on the existing enterprise platform that serves hundreds of large corporations, including Accenture Vanguard Qualcomm, Uberis a. Salesforce.

This distinction is crucial for enterprise adoption. Writer’s system offers complete audit trails that show how the agent arrived at its conclusions and took actions. Consumer AI tools are often “black boxes” without transparency. Shetrit stressed that transparency is essential for regulated industries. “When you’re talking about the largest companies around, whether they are banks, pharmaceutical companies, or healthcare providers, it’s unacceptable to not know how autonomous agents behave and what they’re up to. And you can audit and get full visibility into what’s happening in that box.”

Action Agent is able to connect with more than 600 enterprise tools, which is a significant achievement. The agent uses Model Context Protocolis an emerging standard for AI tool Integration, but Writer has enhanced this with enterprise-grade controls to address security and governance concerns.

In order to bring MCP up to enterprise standards, Writer has worked closely with Amazon Web Services as well as other industry players. Shetrit said, referring to recent MCP implementation issues at companies such as Asana and GitHub.

This company’s approach provides granular control beyond simple permissions. “It is not only by a user. Shetrit explained. “As an IT persona or security persona, i have the controls i need to feel confident with this data access.” Shetrit explained that “you need that fine-grained control and that’s what we’re baking into the system.”

This company has announced support for 600 different tools. Each tool offers fine-grained controls at the integration and agent level. This allows Action Agent’s technology to coordinate across an organization’s entire ecosystem of technology, from customer relationship systems to financial databases.

Free AI agents challenge traditional pricing models

Author’s decision to provide Action Agentis now available for free to existing customers. This represents a challenge to traditional software pricing models, and reflects the broader shifts within the AI industry. The move is made despite the high computational costs associated with token usage by the agent.

Shetrit explained that token pricing is a major problem when it comes to enterprises. “They need to have a budget item. They must determine the cost structure. This highly variable model does not work for them, and we have been moving away for a while. The company spent only $700,000 on training its employees. Palmyra x4model, compared with an estimated $4.6 for a similar sized OpenAI. Writer’s use innovative training techniques and synthetic data to reduce computational requirements is the reason for this efficiency.

The reasoning behind Writer’s free offer goes beyond the company’s competitive positioning. Shetrit stated, “We believe this shows the true value of the platform and ecosystem, and starts to deliver on the promise of AI.” Users have expressed greater excitement about this AI tool than any other AI tool, including copilot systems.

The Enterprise AI market is heating up as startups target Microsoft, Google and other tech giants

The launch of Action Agent intensifies competition in the rapidly growing enterprise AI market. This m arket is projected to grow from 58 billion dollars to $114 billion dollars by 2027. The company competes with Microsoft’s Copilot Suite, Google’s enterprise offerings, and OpenAI business products. However, it targets a different segment of the market with its enterprise-first strategy.

This competitive positioning reflects the industry split between those companies that build general-purpose AI systems, and those that focus specifically on enterprise needs. Writer’s approach puts security, governance and reliability above raw capability in the hope that enterprise customers will opt for specialized tools rather than consumer products adapted to business use. Shetrit said that competitors are focusing on the consumer market, not us. “We are completely on the Enterprise B-to-B side.”

The focus has paid off in terms of financial success. In November 2024, Writer raised $200M in Series C funding at a valuation of $1.9B. This is nearly quadrupling the previous valuation. The round was led by Premji Invest and Radical Ventures with ICONIQ Growth and other major enterprise players such as Salesforce Ventures Adobe Ventures and IBM Ventures.

From automation to transformation: AI will transform corporate work

The writer’s vision goes beyond current automation and fundamentally reshapes how enterprises operate. The company identifies the two clusters of use-cases emerging in large companies: traditional “90% work flow, 10% AI” optimization, and new “90% AI 10% workflow” experiences which unlock new capabilities.

Shetrit predicted that “each employee will have something like this next them, which helps them do their job, helps automate a great deal of it so you can do work with much higher leverage across the organization.”

The transformation aims to address a crucial shift in enterprise software requirements. As employees get used to AI tools in their daily lives, enterprise software needs to match or surpass that quality. Shetrit said, “You can’t afford enterprise software that isn’t as good and, in many cases, even better.”

This shift is already changing the internal dynamics of Writer. “Historically, I have been a PM and can say that execution has always been the bottleneck. So I can say no because I lack capacity. Shetrit explained that capacity is no longer a bottleneck. Shetrit explained that when his product managers say they don’t want to work on projects, he uses Action Agent to generate at least 80%, 70% and 90% of the project for them to start working on.

Inside Writer’s collaboration with Uber

Writer’s collaboration on Action Agent development with Uber illustrates how the technology is improved by leveraging customer relationships. Uber’s AI Solutions provided operational expertise to scale high-quality annotations in complex enterprise domains while validating the agent capabilities through real-world use-cases.

Megha Yethadka is the GM and Head Uber AI Solutions. She said, “Our collaboration allowed us to contribute deep operational expertise in high quality data annotation to shape an agent that can tackle the most complex enterprise problems.” This partnership model allows Writer agents to solve real enterprise problems, rather than hypothetical use cases. The approach has been used in a variety of industries, including HR candidate sourcing, securities analysis, clinical trial site selection, and competitive intelligence.

Shetrit noted how customer creativity continues to amaze the team. “I’m certain, because that’s what platform and technology is, that if we have this discussion again in a few weeks, I’ll come up with completely different use-cases, because our customers are very, very creative.”

What’s next? Rollout timeline

Writer intends to expand Action Agent’s capabilities in the coming weeks. The company will connect to 80 enterprise platforms, as well as third-party data providers such PitchBook and FactSet. This will allow access to the 600+ agent tools.

This rollout strategy is based on lessons learned from enterprise AI deployments. Writer will not launch with all capabilities. Instead, it will start with the core functionality and add integrations over time based on feedback from customers and real-world testing.

Action Agent will be available in beta immediately to Writer’s current customers. A 14-day trial is also available for new users. The company can refine the system using feedback from the enterprise while maintaining the security standards required by regulated industries.

This launch marks a pivotal point in the enterprise AI revolution. Autonomous agents are now mission-critical tools for business, rather than being experimental curiosities. The Writer agent-first approach could determine which companies are able to successfully transition from human-driven work to AI-augmented.

The most telling sign that this shift is underway was Shetrit’s statement during the interview. “We will become managers of this fleet of agents whether they are humans or synthetic agents.”