“Vibe coding” startup Pythagora is looking to take artificial intelligence-powered software development to the next level with the launch of its platform today, saying it will help anyone – including noncoders – to develop full-stack applications with nothing more than a series of prompts. The company says its platform is built for both developers and nontechnical users, and unlike similar generative AI coding tools, unifies both front- and back-end development with comprehensive debugging features to bring the entire app creation experience into a single platform. Pythagora can be thought of as an “AI teammate” that lives inside software development tools such as VS Code and Cursor. It consists of a team of 14 specialized AI agents that can automate various coding-related tasks without supervision, taking care of everything from planning and writing code to testing, debugging and deployment. Pythagora essentially supercharges vibe coding, entirely eliminating the need to actually code. The tool is designed to be less like a coding assistant and more like a co-developer. What that means is it does more than just create the code – it also explains why the code is written as it is, and can walk users through any changes it has made. But users can still intervene and edit the code as they see fit, if they decide it’s necessary to do so.
Google announced its open-source Gemini-CLI that brings natural language command execution directly to developer terminals, offering extensibility architecture, built around the emerging MCP standard
Google announced its open-source Gemini-CLI that brings natural language command execution directly to developer terminals. Beyond natural language, it brings the power of Google’s Gemini Pro 2.5 — and it does it mostly for free. The free tier provides 60 model requests per minute and 1,000 requests per day at no charge, limits that Google deliberately set above typical developer usage patterns. The tool is open source under the Apache 2.0 license. While Gemini CLI is mostly free, OpenAI and Anthropic’s tools are not. Google senior staff software engineer Taylor Mullen noted that many users will not use OpenAI Codex or Claude code for just any task, as it carries a cost. Another key differentiator for Gemini CLI lies in its extensibility architecture, built around the emerging Model Context Protocol (MCP) standard. This approach lets developers connect external services and add new capabilities and positions the tool as a platform rather than a single-purpose application. The extensibility model includes three layers: Built-in MCP server support, bundled extensions that combine MCP servers with configuration files and custom Gemini.md files for project-specific customization. This architecture allows individual developers to tailor their experience while enabling teams to standardize workflows across projects. If an organization wants to run multiple Gemini CLI agents in parallel, or if there are specific policy, governance or data residency requirements, a paid API key comes in. The key could be for access to Google Vertex AI, which provides commercial access to a series of models including, but not limited to, Gemini Pro 2.5. Gemini CLI operates as a local agent with built-in security measures that address common concerns about AI command execution. The system requires explicit user confirmation for each command, with options to “allow once,” “always allow” or deny specific operations. The tool’s security model includes multiple layers of protection. Users can use native macOS Seatbelt support for sandboxing, run the agent in Docker or Podman containers, and route all network traffic through proxies for inspection. The open-source nature under Apache 2.0 licensing allows complete code auditing.
Tray.ai’s platform addresses data incompleteness in AI deployment through integration of smart data sources that simplify synchronization of structured and unstructured enterprise knowledge, ensuring agents are informed with relevant and reliable information
Tray.ai has released Merlin Agent Builder 2.0, a platform designed to address challenges in AI agent deployment within enterprises. The platform aims to bridge the gap between building and actual usage of AI agents, addressing issues such as lack of complete data, session memory limitations, challenges with large language model (LLM) configuration, and rigid deployment options. The updated solution includes advancements in four key areas: integration of smart data sources for rapid knowledge preparation, built-in memory for maintaining context across sessions, multi-LLM support, and streamlined omnichannel deployment. Smart data sources simplify the connection and synchronization of structured and unstructured enterprise knowledge, ensuring agents are informed with relevant and reliable information. Built-in memory capabilities reduce the need for custom solutions and enhance continuity in user exchanges, improving adoption rates. The platform supports multiple LLM providers, allowing teams to assign specific models to individual agents with tailored configurations. Unified deployment across channels allows teams to build an agent once and deploy it seamlessly across communication and application environments, eliminating the need for repeated setup and technical adjustments for different channels. Tray.ai aims to provide a unified platform that enables IT and business teams to transition from pilot projects to production-ready AI agents that are actively used by employees and customers.
Camunda’s agentic orchestration for trade exception management lets clients connect to cloud or on-premises AI models and apply deterministic guardrails offering 86% reduction in manual effort and cutting T+1 delays by 98%
Camunda has highlighted how its agentic orchestration capabilities are enabling organizations to introduce AI at scale into their processes while preserving transparency, compliance, and control. Agentic trade exception management (available on Camunda Marketplace): Camunda’s platform allows clients to connect their preferred AI models, whether hosted in the cloud or internally via EY labs, and apply deterministic guardrails to ensure AI is only triggered when appropriate. This lets clients avoid rebuilding AI from scratch, instead focusing on governance, visibility, and scalable deployment – areas where Camunda’s orchestration brings immediate and measurable value. In one capital markets implementation, EY reduced manual effort by 86%, cut T+1 delays by 98%, and boosted analyst productivity from 6–10 to 41–64 cases per day – a 7x improvement. Agentic AI-assisted quality audit process (available on Camunda Marketplace): Cognizant has created and demonstrated workflows in Camunda that include mandatory human review steps – enabling AI to suggest actions, but requiring manual approval before those actions are executed. This balance allows organizations to benefit from AI-powered insights while also facilitating compliance with regional laws. For example, audit trails, escalation paths, and process visibility are all embedded into the BPMN model, assisting organizations in demonstrating full control over every agentic interaction. This led to significant time savings: the quality audit process was reduced from 138 minutes to just 7–10 minutes, increasing auditor productivity by 20–30%, and cutting costs by 30–50%. All activity is fully traceable via embedded audit trails and escalation paths in BPMN. Customer service agent (available on Camunda Marketplace): Replacing standard auto-responses, Incentro built an AI agent that uses a LLM to analyze queries and draft meaningful replies in real time. The agent accesses the company’s full FAQ and documentation set, enabling specific answers rather than generic acknowledgments. Camunda’s BPMN model structures the logic, with the agent dynamically choosing the best response path via an ad-hoc sub-process. When implementing these systems with Payter, Incentro was able to reduce handling time per inquiry from 24 to 12 minutes, with lead time cut by 58%, helping improve both customer NPS and agent satisfaction without increasing headcount. Compliance agent (available on Camunda Marketplace): BP3 shared how it is integrating agentic AI into decision-heavy workflows in regulated industries like BFSI, pharma, healthcare, and utilities. Its approach uses LLMs alongside DMN (Decision Model and Notation) tables to generate “accept, reject, or refer” outcomes. In ambiguous cases, decisions are escalated to a human, enabling the AI to learn from real-world feedback over time.
Tines workflow automation agents enable enterprises to apply the right level of automation with flexibility from manual to fully autonomous within a single platform and run entirely within the platform’s secure infrastructure
Tines announced autonomous AI capabilities within its workflow automation platform via the launch of agents. Agents mark a significant evolution in Tines’ platform, enabling customers to automate workflows with maximum control and flexibility, whether with deterministic logic, human-in-the-loop copilots, or full AI autonomy. Agents enable Tines customers to build intelligent, context-aware workflows that can act independently, suggest next steps, and collaborate with users in real time. The addition of agents allows customers to choose the right level of AI involvement for every workflow, ensuring organizations can implement AI automation that aligns with their specific security requirements, levels of complexity, and operational needs. Unlike traditional AI implementations that require external data sharing or compromise on security, Tines’ agents run entirely within the platform’s secure infrastructure. This ensures no customer data leaves the environment, is logged, or used for training, delivering the privacy and governance assurances that enterprise teams demand. Tines capabilities: Full-spectrum automation and orchestration: Apply the right level of automation with flexibility—from manual to fully autonomous—within a single platform. Enterprise-grade security: Built by security professionals, Tines keeps all automation and data within its own infrastructure. Seamless system integration: Connect any tool, LLM, or proprietary app to build, augment, and orchestrate intelligent workflows. Intuitive no-code interface: Easily design complex, mission-critical workflows with drag-and-drop tools and built-in collaboration features. User-friendly adoption: Deploy apps, chatbots, and integrations with popular tools such as Slack to boost usage and maximize ROI on AI initiatives.
CC Signals framework will allow creators to publish a document that specifies how AI models may and may not use their content
Creative Commons has previewed an upcoming framework designed to help creators manage how artificial intelligence models use their content. The framework, which is called CC Signals. The new CC Signals framework will allow creators to publish a document that specifies how AI models may and may not use their content. This document will “range in enforceability, legally binding in some cases and normative in others,” Creative Commons staffers noted. The reason is that the extent to which creators can limit AI models’ use of their works varies by jurisdiction. Creative Commons detailed that the framework will include four content usage stipulations. Each one sets forth a different set of requirements for how AI models may interact with a file. CC Signals will allow creators to apply up to two of the four stipulations to a given file. Using CC Signals, creators can also indicate that they expect compensation from AI developers who use their work. That compensation may take several forms. CC Signals will allow for monetary or in-kind contributions to a file’s creators, as well as to the broader “ecosystem from which you are benefiting.” According to Creative Commons, there will be support multiple definitions of open-source AI. Content creators may require that a neural network’s weights be available under a free license. They can also go further and mandate that the tooling with which the algorithm was developed be open source as well.
Qodo launches CLI agent framework that enables developers to create, customize, and deploy their own AI coding agents.
Qodo, maker of an AI coding platform, today announced the release of Qodo Gen CLI, an agent framework that enables developers to create, customize, and deploy their own AI coding agents. With the framework, creating agents can be done by writing configuration files that add autonomous AI agents throughout the software development life cycle, according to the company’s announcement. Qodo was built to help developers add autonomous coding capabilities to their applications without requiring expertise in AI systems, which can lead to solutions that sync up with an organization’s requirements, the company said. With Qodo Gen CLI, developers can define custom agents and what tools they can access, specify actions that trigger the agents, what instructions guide their behavior and ultimately, what their outputs should be. Along with enabling custom agent creation, Qodo Gen CLI includes pre-built agents for code review, test coverage analysis, and release notes generation. These agents integrate seamlessly with existing development tools through GitHub Actions, GitLab CI, Jenkins, and other CI/CD systems. For advanced use cases, agents can be exposed as Model Context Protocol (MCP) servers, enabling integration with other AI tools and platforms.
OpenAI’s API platform allows developers to express intent, not just configure model flows through built-in capabilities for knowledge retrieval, web search, and function calling for supporting real-world agent workflows
Olivier Godement, Head of Product for OpenAI’s API platform, provided a behind-the-scenes look at how enterprise teams are adopting and deploying AI agents at scale. According to Godement, 2025 marks a real shift in how AI is being deployed at scale. With over a million monthly active developers now using OpenAI’s API platform globally, and token usage up 700% year over year, AI is moving beyond experimentation. Godement emphasized that current demand isn’t just about chatbots anymore. “AI use cases are moving from simple Q&A to actually use cases where the application, the agent, can do stuff for you.” This shift prompted OpenAI to launch two major developer-facing tools in March: the Responses API and the Agents SDK. Some enterprise use cases are already delivering measurable gains. Godement positioned the Responses API as a foundational evolution in developer tooling. Previously, developers manually orchestrated sequences of model calls. Now, that orchestration is handled internally. “The Responses API is probably the biggest new layer of abstraction we introduced since pretty much GPT-3.” It allows developers to express intent, not just configure model flows. “You care about returning a really good response to the customer… the Response API essentially handles that loop.” It also includes built-in capabilities for knowledge retrieval, web search, and function calling—tools that enterprises need for real-world agent workflows. Some enterprise use cases are already delivering measurable gains: Stripe, which uses agents to accelerate invoice handling, reporting “35% faster invoice resolution; ” Box, which launched knowledge assistants that enable “zero-touch ticket triage.” Other high-value use cases include customer support (including voice), internal governance, and knowledge assistants for navigating dense documentation. Godement offered a glimpse into the roadmap. OpenAI is actively working on: Multimodal agents that can interact via text, voice, images, and structured data; Long-term memory for retaining knowledge across sessions; Cross-cloud orchestration to support complex, distributed IT environments. What matters now is building a focused use case, empowering cross-functional teams, and being ready to iterate. The next phase of value creation lies not in novel demos—but in durable systems, shaped by real-world needs and the operational discipline to make them reliable.
Study finds running gen AI models on the phones instead of in the cloud consumed anywhere from 75% to 95% less power, with associated sharp decreases in water consumption and overall carbon footprint
One of the easiest ways to minimize AI’s environmental impact may be to move where the processing is done, per new academic research conducted in partnership with Qualcomm. Running AI on devices instead of in the cloud slashes power consumption of queries by about 90%, the study finds. The industry has long touted the benefits of running models locally on devices instead of in the cloud — not just in energy terms, but also potentially making them cheaper and more private. Researchers at the University of California, Riverside ran a series of experiments comparing the performance of various generative AI models, both in the cloud and on phones powered with Qualcomm chips. Running any of six different models on the phones consumed anywhere from 75% to 95% less power, with associated sharp decreases in water consumption and overall carbon footprint. Qualcomm is also developing an AI simulator and calculator that illustrates, for any given query and user location, what the responses would look like on-device versus the cloud, and how much less power and water they would use. One example — running a coding skills question on the Llama-2-7B model in California — was 94% more power efficient and 96% more water efficient on-device. For all six models in the study, the inference time on the phones, measured in seconds, was higher than in the cloud. Narrowing or eliminating that gap, particularly on the most powerful and popular models, will be crucial to accelerating on-device adoption. For many AI users, the data center in your pocket might be all you need.
SAP Fioneer’s AI agent allows finance teams to generate complex reports using natural language by leveraging the suspense account analysis without the need to share data externally
SAP Fioneer has launched its AI Agent: an expert-built solution designed to intelligently enhance core operations of financial services institutions. By leveraging the suspense account analysis, finance teams can generate complex reports using natural language, significantly reducing manual effort, improving operational efficiency, and achieving considerable time savings. The Fioneer AI Agent delivers intelligence that is integrated into SAP Fioneer’s banking, insurance, and finance solutions, offering contextual, transparent, and actionable use cases without the need for custom development and heavy IT dependance. It empowers financial professionals to interact with data using natural language, eliminating reliance on IT teams and accelerating time to value. Designed for flexibility, the Fioneer AI Agent supports bring-your-own-LLM strategies as well as SAP BTP AI Core LLMs and will integrate with SAP Joule and other agents such as Microsoft Copilot. Integrated and aligned with the SAP strategy, it ensures full compliance with data privacy and auditability standards, making it a trusted solution for institutions seeking to scale AI responsibly and effectively. The first release of the Fioneer AI Agent lays the foundation for banks and insurers to automate processes, gain real-time insights, and make smarter decisions using natural language and without the need to share data externally. The Fioneer AI Agent is generally available now as an add-on for SAP Fioneer S/4HANA products in Banking, Insurance, and Finance.