Perplexity Launches $200/Month AI Agent Bundle With ChatGPT, Gemini, and Grok

Perplexity unveils Perplexity Computer, a general-purpose AI system that orchestrates multiple frontier models to run complex workflows autonomously.

By Daniel Mercer Edited by Maria Konash Published: Updated:
Perplexity unveils Perplexity Computer, an AI system that coordinates long-running, cross-platform workflows using multiple models. Photo: Perplexity

Perplexity AI has introduced Perplexity Computer, a new system designed to unify frontier AI models, including ChatGPT, Gemini, and Grok, into a single, general-purpose “digital worker” capable of creating and executing entire workflows.

The company argues that while AI models are becoming increasingly powerful, the interfaces built around them are now the bottleneck. Perplexity Computer aims to remove that limitation by moving beyond chat interfaces and task-based agents toward a system that can autonomously design, coordinate, and run multi-step processes for hours — or even months.

From Answers to Autonomous Workflows

Unlike traditional chat-based AI systems that generate responses, Perplexity Computer begins with an outcome. Users describe a goal, and the system decomposes it into tasks and subtasks, automatically spawning specialized sub-agents to execute them.

These sub-agents can perform web research, generate documents, process data, make API calls to connected services, and even write code. Tasks are coordinated asynchronously: one agent can gather data while another drafts a report. If problems arise, the system creates additional sub-agents to troubleshoot — whether that means researching documentation, locating API keys, building small apps, or escalating only when necessary.

Each task runs inside an isolated compute environment with access to a real filesystem, browser, and tool integrations, creating what Perplexity describes as a secure, universal harness for advanced AI work.

Multi-Model by Design

Perplexity emphasizes that its system is model-agnostic and built around intelligent multi-model orchestration. Rather than relying on a single foundation model, Perplexity Computer dynamically assigns tasks to the most suitable AI system.

As of launch, the platform runs Opus 4.6 as its core reasoning engine and deploys other frontier models depending on the job — Gemini for deep research and sub-agent creation, Nano Banana for image generation, Veo 3.1 for video, Grok for lightweight speed-focused tasks, and ChatGPT 5.2 for long-context recall and broad search.

The company argues that, contrary to claims that AI models are commoditizing, they are in fact specializing. In that environment, the most powerful system is not a single model, but an orchestrator that intelligently combines them.

A Broader Evolution

Perplexity frames the launch as a continuation of its broader mission to “power the world’s curiosity.” Previous steps included Comet, described as an AI-native browser, and Comet Assistant, a personal AI agent. With deep research capabilities, persistent memory, and task management already in place, Perplexity Computer represents the next step: AI not just as an assistant, but as an operational system.

The company draws a historical parallel to 18th-century “computers” — human apprentices who performed complex mathematical calculations collaboratively. In that sense, Perplexity argues, the word has come full circle: AI is now the computer.

Perplexity Computer is available immediately to Perplexity Max subscribers at $200 per month, with Enterprise Max access expected soon.

AI Is Silently Making Cybersecurity Talent More Valuable Than Ever

New AI cybersecurity systems like Anthropic’s Project Glasswing could increase demand for security professionals as threats and vulnerabilities scale faster.

By Marcus Lee Edited by Maria Konash Published:
AI cybersecurity tools like Glasswing drive demand for security talent as threats grow. Image: Philipp Katzenberger / Unsplash

Anthropic’s recent launch of Project Glasswing and its experimental Claude Mythos model is triggering debate across the cybersecurity industry, with some experts warning it could significantly reshape both threat detection and workforce demand. Tal Hoffman, founder of EnclaveAI, said the shift could sharply increase demand for cybersecurity professionals as AI-driven tools expose more vulnerabilities at scale. While specific claims about the model’s capabilities remain under scrutiny, the broader direction points to a move toward AI systems capable of autonomously identifying and validating software weaknesses.

Project Glasswing brings together major industry players, including Amazon Web Services, Google, Microsoft, and CrowdStrike, to apply advanced AI models to defensive cybersecurity. Early reports suggest such systems can uncover high-severity vulnerabilities in mature codebases and, crucially, demonstrate whether those flaws are exploitable. This ability to move from detection to validation marks a significant shift in how security risks are assessed.

From Discovery to Exploitation

Traditionally, vulnerability scanning has been plagued by false positives, requiring extensive manual triage by security teams. AI-driven systems promise to improve signal quality by surfacing fewer but more meaningful findings. More importantly, they can potentially automate exploit validation, reducing the gap between identifying a flaw and proving it can be used in an attack.

This shift could dramatically increase the volume of actionable vulnerabilities. While automation may reduce some manual work, it also creates a new bottleneck: remediation. Security teams may face a surge in verified issues that require prioritization, architectural decisions, and fixes—tasks that still depend heavily on human expertise.

Growing Asymmetry in Cyber Defense

The controlled rollout of these tools highlights another emerging challenge: uneven access. Anthropic has restricted availability of its most advanced models to a limited group of organizations, citing safety concerns. While this approach aligns with responsible deployment practices, it creates a gap between companies with access to advanced AI defenses and those without.

At the same time, the attack surface is expanding rapidly. AI agents, internal automation tools, and integrations across business functions are introducing new vulnerabilities, often without formal security review. These developments are creating new categories of risk, including prompt injection attacks and insecure AI-driven workflows.

Rising Demand for Security Talent

Despite advances in automation, experts argue that demand for cybersecurity professionals is likely to increase rather than decline. As AI systems surface more vulnerabilities and accelerate workflows, organizations will need more specialists to interpret findings, implement fixes, and manage evolving threat models.

The role of security teams is shifting from finding vulnerabilities to managing and mitigating them at scale. This transition could elevate cybersecurity from a back-office function to a strategic priority, particularly as AI-driven risks begin to impact critical infrastructure and enterprise systems.

In the longer term, AI-powered tools may strengthen defenders by enabling continuous monitoring and faster response times. However, in the near term, the gap between emerging capabilities and widespread access remains a key challenge. Organizations that invest early in AI-driven security tools and talent may be better positioned to navigate this transition as the cybersecurity landscape evolves.

AI & Machine Learning, Cybersecurity & Privacy, News

Claude Code Gets ‘Routines’ to Enable Autonomous AI Tasks

Anthropic has introduced “Routines” in Claude Code, enabling autonomous AI workflows triggered by schedules, APIs, or GitHub events.

By Daniel Mercer Edited by Maria Konash Published:
Anthropic launches Claude Code Routines for trigger-based, autonomous AI workflows. Image: Claude Code

Anthropic has introduced a new feature called “Routines” to its Claude Code platform, marking a step toward fully autonomous AI workflows that can operate continuously in the background. The feature, currently in research preview, allows users to define tasks that run automatically based on triggers such as schedules, API calls, or GitHub events.

Routines effectively transform Claude Code from an interactive coding assistant into a persistent, cloud-based agent. Once configured, a routine can execute tasks without user intervention, even when a device is offline. Each routine combines a prompt, connected repositories, and external integrations into a reusable workflow that can be triggered repeatedly.

The system supports multiple trigger types. Scheduled routines can run at regular intervals such as hourly or weekly, while API-triggered routines can be activated programmatically via HTTP requests. GitHub-based triggers allow the system to respond automatically to events like pull requests or releases. These triggers can be combined, enabling more complex automation scenarios.

From Assistant to Autonomous Agent

The introduction of Routines reflects a broader shift in AI tools from reactive assistants to proactive agents. Instead of responding to prompts, Claude Code can now initiate actions such as reviewing code, triaging alerts, updating documentation, or managing workflows across tools like Slack and GitHub.

Example use cases include automated code reviews, incident response workflows, backlog management, and deployment verification. In these scenarios, Claude can analyze data, generate outputs such as pull requests, and communicate results without human input, leaving users to review outcomes rather than perform repetitive tasks.

Routines run as full cloud sessions with access to selected repositories, environments, and connectors. This allows them to execute shell commands, interact with external services, and modify codebases, depending on permissions. However, the feature also introduces new considerations around security and governance, as actions are performed under the user’s identity and can affect production systems.

Expanding the Agent Ecosystem

The launch comes as Anthropic continues to expand its developer-focused offerings and compete with other AI platforms in building agent-based systems. Routines are available across Pro, Team, and Enterprise plans, signaling a push toward enterprise adoption where automation and integration are key requirements.

The update also arrives alongside expectations of a new model release, widely anticipated to further enhance Claude’s reasoning and coding capabilities. While details remain limited, the combination of more capable models and autonomous execution features points to a future where AI systems handle increasingly complex workflows end-to-end.

With Routines, Anthropic is positioning Claude Code not just as a tool for developers, but as an infrastructure layer for automated work. As organizations experiment with these capabilities, the balance between efficiency gains and operational risk will likely shape how quickly such systems are adopted at scale.

Half of Americans Now Use AI Weekly, ChatGPT Leads

A new national poll shows 50% of Americans used AI in the past week, with ChatGPT leading adoption across work, learning, and creative tasks.

By Samantha Reed Edited by Maria Konash Published:
Poll finds half of Americans use AI weekly, with ChatGPT leading across work and creativity. Image: Eyestetix Studio / Unsplash

A new national poll from Epoch AI and Ipsos finds that artificial intelligence has reached mainstream adoption in the United States, with 50% of adults reporting they used an AI service in the past week. The data highlights how quickly AI tools have moved from niche technology to everyday utility, with ChatGPT emerging as the most widely used platform.

According to the survey, 31% of Americans reported using ChatGPT in the past week, ahead of competitors such as Google Gemini (21%), Microsoft Copilot (11%), and Meta AI (8%). Usage is also frequent, with 65% of AI users engaging with these tools multiple days per week and 16% using them nearly every day.

The findings show that AI is being applied across a wide range of tasks. Around 80% of users rely on AI for information lookup or recommendations, while 59% use it for writing and editing, 55% for learning or advice, and 53% for brainstorming ideas. More advanced use cases are also gaining traction, including image generation (44%) and data analysis or programming (37%).

AI Becomes a Daily Tool

The poll suggests AI is becoming embedded in everyday digital workflows. Most users interact with AI by typing prompts directly, but many also rely on integrated experiences such as AI-powered search summaries or built-in assistants within productivity software. For example, a majority of Copilot users access AI within tools like Word, Excel, or Teams, while nearly half of Gemini users encounter AI-generated summaries in search results.

This integration is accelerating adoption by reducing friction. Rather than seeking out standalone tools, users increasingly encounter AI as part of the platforms they already use, making it a default layer in digital interactions.

Impact on Work and Productivity

The survey also highlights AI’s growing role in the workplace. Among employed respondents who use AI, 51% report using it for work-related tasks. Within that group, 26% primarily use AI for work, while another 25% split usage evenly between professional and personal purposes.

AI is already reshaping job responsibilities. One in five workers said AI now performs tasks they previously handled themselves, while 15% reported taking on new responsibilities enabled by AI tools. Despite this, access remains uneven: half of workers using AI rely on personal accounts or free versions, while only one-third use tools provided by their employer.

The results point to a transitional phase in enterprise adoption. While individuals are rapidly integrating AI into their workflows, many organizations are still formalizing policies, infrastructure, and access. As companies catch up, AI usage is likely to become more standardized across workplaces.

Overall, the data underscores a shift from experimentation to routine use. With half of Americans already engaging with AI weekly, tools like ChatGPT and its competitors are becoming a foundational layer of modern work and everyday decision-making.

AI & Machine Learning, Consumer Tech, News

AWS Launches Amazon Bio Discovery to Accelerate Drug Design

AWS has launched Amazon Bio Discovery, an AI-powered platform that helps scientists design, test, and refine drugs faster using integrated models and lab workflows.

By Laura Bennett Edited by Maria Konash Published:

Amazon Web Services has launched Amazon Bio Discovery, a new AI-powered application designed to help scientists accelerate drug discovery by combining machine learning models with real-world lab testing. The platform introduces a “lab-in-the-loop” workflow, where AI-generated drug candidates are tested experimentally and fed back into the system to improve future results.

The application provides access to a broad catalog of biological foundation models, or bioFMs, trained on large biological datasets. These models can generate and evaluate potential drug candidates, particularly antibodies, during early-stage research. Scientists interact with the system through an AI agent that helps design experiments, select appropriate models, and optimize inputs using natural language rather than code.

Amazon Bio Discovery is designed to lower barriers to AI adoption in life sciences. Traditionally, using advanced models required specialized computational expertise and infrastructure. The new platform simplifies this process by offering pre-benchmarked models, automated workflows, and integrated tools for comparing performance. Researchers can also fine-tune models using their own experimental data without building custom pipelines, keeping proprietary data secure within their organization.

Closing the Loop Between AI and the Lab

A key feature of the platform is its integration with laboratory partners, including Twist Bioscience and Ginkgo Bioworks. Scientists can send AI-generated candidates directly for synthesis and testing, with results automatically routed back into the system. This creates a continuous feedback loop, allowing each experiment to improve the next iteration.

The approach has already shown early results. In collaboration with Memorial Sloan Kettering Cancer Center, researchers used the platform to design hundreds of thousands of antibody candidates for pediatric cancer therapies. What traditionally takes months or even a year was reduced to a matter of weeks, from initial design to lab testing.

Democratizing AI in Life Sciences

Amazon Bio Discovery reflects a broader push to make advanced AI tools accessible to a wider range of scientists, not just those with machine learning expertise. By combining model access, experiment design, and lab coordination into a single platform, AWS aims to streamline workflows that are often fragmented across multiple systems and teams.

The platform is built on infrastructure already widely used in the pharmaceutical industry, with AWS noting that 19 of the top 20 global drugmakers rely on its cloud services. Early adopters include Bayer, the Broad Institute, and Fred Hutch Cancer Center. The launch also aligns with a wider wave of AI-driven partnerships across the sector, such as Novo Nordisk teaming up with OpenAI to accelerate drug discovery for obesity and diabetes treatments.

As AI becomes more embedded in drug development, platforms like Amazon Bio Discovery highlight a shift toward integrated systems that connect computational design with real-world experimentation. This convergence could significantly shorten development timelines and expand access to advanced research tools across the life sciences ecosystem.

AI & Machine Learning, News, Research & Innovation

OpenAI Buys Hiro Finance in Strategic Talent Acquisition

OpenAI has acquired personal finance startup Hiro Finance in an apparent acquihire, bringing its team and expertise into its growing AI ecosystem.

By Samantha Reed Edited by Maria Konash Published: Updated:
OpenAI acquires Hiro Finance in acquihire, adding fintech talent to expand AI in financial tools. Image: Hiro Finance

OpenAI has acquired personal finance startup Hiro Finance in what appears to be a talent-focused deal, as the company continues expanding its capabilities across business and consumer applications. Financial terms were not disclosed, but Hiro will shut down operations on April 20 and delete user data by May 13, indicating a full integration into OpenAI.

Hiro was founded in 2023 by Ethan Bloch and developed an AI-powered financial planning tool designed to help users model different financial scenarios. The app allowed consumers to input data such as income, expenses, and debt, and then simulate outcomes to guide decision-making. The startup positioned itself around accuracy in financial calculations, addressing a longstanding weakness in earlier AI systems.

As part of the acquisition, Hiro’s team will join OpenAI, though the exact number of employees has not been disclosed. The company was backed by prominent venture firms including Ribbit Capital, General Catalyst, and Restive Ventures. The move suggests OpenAI is prioritizing talent and domain expertise as it builds out specialized AI applications.

Expanding Into Financial Workflows

The acquisition highlights OpenAI’s growing interest in financial use cases. Its flagship products are already widely used by finance teams for analysis, reporting, and forecasting. Adding Hiro’s expertise could strengthen OpenAI’s ability to deliver more tailored tools for both consumers and enterprises.

This is not OpenAI’s first move in the financial space, and it reflects a broader trend of AI companies targeting high-value professional workflows. Financial planning, in particular, offers a compelling use case due to its reliance on data modeling, projections, and scenario analysis—areas where AI models have improved significantly in recent years.

Bloch brings prior experience in fintech, having previously founded Digit, a digital banking service that was acquired for over $200 million. His background in consumer finance products may help OpenAI explore new applications or refine existing offerings in this domain.

Talent as a Strategic Asset

The deal underscores a common pattern in the AI industry: acquisitions driven more by talent than by standalone products. With Hiro shutting down shortly after the acquisition, the focus appears to be on integrating its team and expertise into OpenAI’s broader roadmap.

The move also comes amid intensifying competition in AI, particularly in areas like coding agents and enterprise tools. By bringing in specialized teams, OpenAI can accelerate development in targeted domains without building capabilities from scratch.

Exit mobile version