When Intuit launched AI agents to serve 3 million customers, 85% of users returned. According to Marianna Tessel, EVP and GM at Intuit, the key was blending AI technology with human intelligence to build trust and confidence. This hybrid approach, combining AI agents with access to human experts in accounting, tax, and payroll, proved more effective than using chatbots alone. The company’s GenOS platform and Intuit Intelligence dashboard help automate tasks like invoicing, payroll, fraud detection, and reporting, cutting manual effort by 30% and accelerating payments. Users can consult human specialists for complex issues or high-stakes decisions, enhancing reliability. Intuit continues to innovate with user-friendly tools that allow business owners to customize AI-driven workflows without needing coding skills. This blend of human insight and AI technology sets a new standard for enterprise AI adoption.
ThinkLabs AI, backed by Nvidia and other strategic investors, has raised $28 million in Series A funding to accelerate its development of AI models that simulate electric grid behavior in real-time. The company's physics-informed AI compresses complex power flow analyses from weeks to minutes with over 99.7% accuracy, enabling utilities to adapt quickly to surging electricity demand driven by data centers and electrification. Led by CEO Josh Wong, ThinkLabs differentiates itself by performing detailed three-phase power flow analysis, crucial for utility decision-making on infrastructure investments and operational strategies. Partnerships with Nvidia, Edison International, and Microsoft enhance its platform's capabilities and integration with existing utility ecosystems. Despite conservative utility buying cycles, ThinkLabs is rapidly expanding its customer base and aims to modernize grid planning, addressing workforce renewal and energy affordability challenges. Wong envisions AI as the navigation system for the grid, optimizing resource use rather than building more infrastructure, with the next few years being pivotal for AI's role in shaping energy future.
Softr, a Berlin-based no-code platform trusted by over one million users and companies like Netflix, Google, and Stripe, has introduced an AI-native platform designed to create production-ready business applications. Their new AI Co-Builder enables non-technical users to articulate their software needs in plain language, with the platform instantly generating fully integrated apps including databases, user permissions, and business logic. Unlike many AI-driven app builders that produce unreliable prototypes, Softr’s approach relies on proven building blocks that ensure security and stability, while allowing ongoing no-code visual editing. The platform, which has evolved from a simple Airtable interface to support multiple data sources, serves a broad market ranging from tech giants to non-tech companies in real estate and logistics. Profitable and growing with an organic product-led approach, Softr aims to empower billions of business users to create custom software solutions easily, positioning itself uniquely between traditional no-code tools and AI code generation startups.
In today’s enterprises, team communication often turns into "coordination theater," where more time is spent talking than doing. Tools like Slack and Teams enable rapid discussions but don’t provide the consistent, secure context AI agents need, leading to inefficiencies such as AI hallucinations or repeated clarifications.
PromptQL, a spin-off from the GraphQL company Hasura, is reshaping this dynamic. It converts everyday team messages into a secure, living internal wiki — a persistent source of actionable, proprietary data that evolves alongside your organization. This AI-native workspace automatically tags, assigns, and documents tasks without manual input, turning conversations into immediate, executable actions.
PromptQL’s key innovation is its Shared Wiki, which accumulates context as teams work, preventing AI memory loss or outdated data errors. This wiki organically connects knowledge "islands" across tools like Salesforce, Snowflake, and Zendesk. Human oversight ensures only valid info is added. The virtual SQL layer queries data live without replication, enhancing security and accuracy.
Supporting a broad range of AI models and workflows, PromptQL allows seamless delegation to specialized agents like Claude Code. For example, it can detect and fix code issues in real time, update documentation, and generate reports by connecting disparate data—all within familiar chat threads.
Security is paramount, with fine-grained access control and approval processes guarding sensitive actions. Pricing is usage-based, rewarding broad adoption without penalizing team size. Enterprises benefit from dedicated infrastructure options that guarantee data sovereignty.
As organizations seek AI that acts securely and reliably, PromptQL replaces static chat tools with a collaborative environment where AI agents and humans work side by side, accelerating productivity while maintaining rigorous compliance. This platform paves the way for actionable AI adoption at scale, eliminating coordination inefficiencies and unlocking new levels of team effectiveness.
Slack has introduced more than 30 advanced features to Slackbot, transforming it from a simple AI assistant into a comprehensive enterprise digital coworker. These capabilities include taking meeting notes across various video platforms, working beyond the Slack interface, integrating with thousands of third-party apps via the Model Context Protocol, and acting as a lightweight CRM for small businesses—all without additional installations. Developed in partnership with Anthropic and powered by the Claude model, Slackbot leverages deep context from Slack conversations to deliver tailored, efficient assistance while maintaining user privacy through opt-in controls. This update positions Slackbot as a key part of Salesforce's vision to create an "agentic operating system" that seamlessly connects AI agents, enterprise applications, and users. Despite competitive pressure from Microsoft and Google, Slack bets on its unique context advantages and user experience to drive adoption. The update also raises important questions about privacy and workplace surveillance, which Slack addresses with strict user control features. This launch marks a pivotal moment under interim CEO Rob Seaman, as Slack aims to retain its simplicity amidst growing complexity.
Last week, a product manager on our team built and shipped a feature in just one day — without a spec, ticket, or handoff. Similarly, our designer recently fixed a UI issue directly, avoiding the traditional, slow process of screenshots, ticketing, and sprints. With AI automating scaffolding, tests, and repetitive coding tasks since 2025, engineering bottlenecks have shifted to decision-making and coordination processes, which now slow down software delivery more than implementation itself. This shift means that those closest to the product vision—product managers and designers—can build and ship software directly, eliminating layers of process and delays. As implementation costs approach zero, the old organizational structures and workflows geared to protect engineering bandwidth become obsolete. This new dynamic accelerates feedback loops, sharpens specifications, and encourages ownership across teams. Ultimately, this shift signals a fundamental change in software organizations: everyone, not just engineers, can and will ship software.
Many users have tried AI tools without seeing much benefit, but my real-world experience tells a different story. Over the past six months, I transformed my engineering team to prioritize AI-first workflows, significantly accelerating our development speed. Despite reducing our team size from 36 to 30, our output increased by 170%, roughly doubling our efficiency.
We shifted from lengthy design phases to rapid prototyping, using AI to quickly turn ideas into viable products. This approach not only sped up development but also enhanced product quality and customer satisfaction by integrating AI-powered testing directly into our workflows. Our QA team evolved from traditional testers into system architects who build AI agents to automate acceptance tests, fundamentally changing how we validate and ensure software quality.
On a structural level, software development has inverted from a linear assembly line to a control-tower model, where humans set intent and validate outcomes, while AI handles execution efficiently. Our engineers now operate at a higher abstraction level, focusing on orchestrating AI workflows rather than traditional coding, marking a new era where strategic thinking and AI collaboration drive innovation.
The enterprise voice AI sector is rapidly evolving with major players pushing innovations. Mistral AI, based in Paris, just introduced Voxtral TTS, an open-weight text-to-speech model designed for enterprise use, distinctively allowing companies to download and run it independently without relying on third-party services. This contrasts with the proprietary subscription models of competitors, such as ElevenLabs. Voxtral TTS features a compact 3.4-billion-parameter architecture, achieving six times real-time speech speed, supporting nine languages, and excelling in custom voice adaptation, including zero-shot multilingual capabilities. Independent tests show a preference for Voxtral over ElevenLabs voices in customization tasks. Mistral's approach emphasizes ownership and control over voice AI, addressing data sovereignty concerns especially relevant in Europe. Voxtral TTS completes Mistral's enterprise AI stack, enabling seamless voice-agent applications across industries. The company plans to expand language and dialect support while advancing towards fully end-to-end audio understanding models, aiming to deliver richly expressive, efficient, and controllable AI-driven voice interfaces.
Intercom, a well-established customer service platform, recently unveiled Fin Apex 1.0, an AI model specifically designed to enhance customer support. This model reportedly outperforms leading AI models like GPT-5.4 and Claude Sonnet 4.6, achieving a 73.1% resolution rate in customer issues without human intervention, which surpasses its competitors by about 2%. Besides accuracy, Fin Apex is faster and reduces hallucinations significantly compared to others. Uniquely, it operates at about one-fifth the cost of typical frontier models and integrates seamlessly into Intercom's existing pricing structure. Although Intercom has withheld details on the base model used, citing strategic reasons and plans for evolution, the firm emphasizes that their key advantage lies in post-training on proprietary customer data. This approach tailors the AI to meet specific domain needs, enhancing its effectiveness. The company also highlights the model's role in drastically improving customer experience, shifting the focus from cost savings to quality interactions. As Fin grows rapidly, expected to make up half of Intercom’s revenue soon, the model currently powers over two million weekly customer interactions and is exclusive to Intercom’s services, not available for independent licensing or API use. Intercom plans to expand the use of Fin Apex into sales and marketing sectors, competing directly with solutions like Salesforce's AI agent initiatives.
Traditional Retrieval-Augmented Generation (RAG) systems struggle with maintaining coherence in long-term, multi-session AI interactions, leading to inefficient token use and context overload. xMemory, developed by researchers at King’s College London and The Alan Turing Institute, offers a novel hierarchical memory organization that structures conversations into semantic themes, significantly reducing redundant token consumption. This four-level hierarchy—from raw messages to episodes, semantics, and themes—enables targeted, uncertainty-driven retrieval that cuts token usage nearly in half while improving reasoning accuracy. Unlike conventional RAG, which falters in conversational memory due to correlated, repetitive content and risky pruning, xMemory's top-down search ensures meaningful, non-redundant recall tailored to user queries. Though it requires more upfront processing, xMemory’s approach benefits enterprise applications needing reliable, context-aware AI over extended periods, such as personalized support or coaching, while simplifying retrieval costs and latency during answer generation. The system’s code is openly available for commercial use, and its foundational innovation lies in carefully decomposing and indexing memory rather than focusing solely on retrieval prompts. As AI use cases grow, future challenges will include memory lifecycle management and privacy across collaborative agents.
