Practical AI workflows for knowledge workers - what's new and how to actually use it.

Gemini Spark has entered the competitive landscape of agentic consumer products. Comparison of meeting transcript analysis across three products: Gemini Spark, Claude Co-work, and Chat GPT. A prompt extraction method is used to evaluate action plans from meeting transcripts.

ChatGPT Work is a new mode focused on executing tasks without needing coding skills. Users can build live websites quickly by defining an outcome and using specific tools. A paid subscription may be necessary for advanced features, especially website building.

Comparison of AI tools ChatGPT, Claude, and Gemini in real-world applications. Focus on four key use cases: meeting summaries, business context analysis, customer support, and task management. ChatGPT outperforms others in detail and thoroughness, while Claude excels in concise, human-sounding output.

OpenAI has revamped ChatGPT, introducing a new work tab and mobile app. New models: Saul, Terra, and Luna aim to enhance functionality. A unique 'sites' feature allows users to create websites easily. The free plan offers many updates that competitors lack. Improved plugin integration enhances project workflows.

Fable 5 is back, along with Sonnet 5 and Opus 4.8. Fable 5 is the most capable model, but has tight usage limits. Sonnet 5 is faster and cheaper than Opus, making it the default choice. Fable 5 excels at organizing chaotic information. Limited free access to Fable 5 ends soon; users are encouraged to use its capabilities.

Codex offers a record and replay feature for workflows. Users can capture their actions and audio for later reuse. Skills created can be scheduled for future execution.

OpenAI Codex introduces a record and replay feature for workflows. The new functionality is currently Mac-only and requires a subscription. Comparisons show that Codex's performance surpasses similar functionalities in Claude.

Long press the send icon to access intensity settings. Adjusting intensity affects response quality and cost. Higher intensity yields better responses for complex queries.

ChatGPT now allows users to send emails directly. The feature requires enabling the Gmail connector. Users can create and send emails smoothly within the chat interface.

AI usage is limited by tokens, not messages. Understanding tokens can maximize your AI plan's efficiency. Use the appropriate model for your task to conserve tokens.

ChatGPT now features job search and resume customization tools. Users can generate job listings based on specific resumes. The resume creation upgrade is globally accessible, while job search is US only. Interacting with ChatGPT effectively can yield real job applications and tailored resumes.

ChatGPT now includes a scheduled tasks feature. Users can set up daily briefings with presets in the new schedule tab. This functionality enhances convenience and showcases advanced AI capabilities.

ChatGPT now supports scheduling tasks effectively. The Gmail connector can send emails directly. New feature allows interactive visual aids for generated content. Recent updates improve the customization of scheduled tasks. A competitive Chinese model is comparable to GPT 5.5.

Apple announces an improved version of Siri is coming. Partnership with Google Gemini aims to enhance Siri's capabilities. Developer beta to be released in July.

Cloud users receive double usage for Co-work next month. Increased capacity allows for more scheduled tasks. Applies to all subscription tiers from $20 to $200.

Anthropic's Claude Fable 5 is the best AI model for agentic and knowledge work tasks. It achieves over 80% on software engineering benchmarks, significantly outperforming previous models. User reviews describe it as a mature and efficient programming assistant. The model's high cost requires a subscription and is more expensive than Claude Opus 4.8. Fable 5 has shown its efficacy in real-world applications, such as performing security audits.

OpenAI is merging Codex into the ChatGPT app. The new ChatGPT desktop app will automate tasks and integrate plugins. Expected improvements are projected within six months.

Transform Claude into a visual learning assistant with a specific prompt. Use a single prompt to get multiple visual options before selecting the best one. Learn the difference between interactive visuals and artifacts for sharing and collaboration.

OpenAI upgrades ChatGPT's memory system. New memory summary section allows for easier management. Users can correct inaccuracies directly in memory. Feature similar to Claude's reporting format. Updates made will influence future chat interactions.

Both ChatGPT and Claude have free and paid plans, but ChatGPT offers more generous usage limits. Claude provides better quality writing out of the box, resembling natural human language more closely. Voice input capabilities favor ChatGPT, offering a significantly better and more reliable experience. ChatGPT excels in image generation, featuring the best tool currently available, while Claude lacks this capability.

OpenAI upgraded ChatGPT's memory system significantly. New memory format is more comprehensive and user-friendly. ChatGPT will merge features with the CodeX app for improved performance. Users need to manage their memory settings for optimal results. Comparison made with Claude shows areas for improvement in ChatGPT's memory.

Claude Code can generate hundreds of AI agents to perform various tasks. The workflow feature allows for the automation of complex tasks like creating a personal finance dashboard. The Opus 4.8 model provides extensive context handling for better execution. Claude Code thoroughly plans, verifies, and tests its own work.

Google is personalizing search results based on user preferences. AI overviews at the top of search results will differ for each user. User feedback indicates mixed reactions to these changes.

Introduction of Opus 4.8 by Anthropic. Available in Claude Cowork, Claude Code, and the API. Capable of creating visually impressive web designs.

ChatGPT can now be used directly in PowerPoint for free. It offers four main features: build, update, understand, and polish. All generated content remains fully editable within PowerPoint.

Claude Opus 4.8 is a significant improvement over 4.7, optimizing ambiguity interpretation. Dynamic workflows allow the spawning of numerous sub-agents for complex tasks, impacting cloud resource usage. User feedback on Opus 4.8 is generally positive, with improved benchmarks over previous models.

Google Gemini excels in video understanding compared to competitors. Uploaded a 27-minute video to Gemini, which summarized it quickly in under a minute. Chat version took 9 minutes with additional tools but still managed to summarize. Claude cannot process videos at all. For video-related tasks, Google is currently the leading platform.

Google's search is now more conversational with an AI mode integration. Users can upload images and files directly into search for analysis. The search results can engage in ongoing conversations for more tailored responses. Google Personal Intelligence connects search to personal apps like Gmail and Google Photos. More features, such as agents and proactive notifications, are set to roll out in the coming months.

Google released the Gemini 3.5 Flash, focusing on speed and efficiency. Gemini offers the best free tier among AI providers, with generous usage limits. The $100 Ultra plan bundles additional services, enhancing its value proposition.

Claude offers new plugins tailored for small businesses. Plugins integrate with QuickBooks, PayPal, Stripe, and HubSpot. Key features include payroll planning and invoice tracking.

Claude Code allows users to set specific goals. The AI works independently to achieve these goals. Users experience minimal interaction during the process.

Anthropic introduces a mission control center for Claude Code. Users can manage multiple agents performing different tasks simultaneously. Improved interfaces aim to make interactions more intuitive.

New prompting strategies for AI models have emerged. Shorter, outcome-first prompts are now more effective. Reevaluate existing prompts for better AI performance.

Claude released new preset packs for entrepreneurs and legal tasks. Plugins combine prompts, connectors, and automations to streamline tasks. The small business plugin pack includes 31 skills for various business operations. Most plugins require human intervention for final execution. Integration with specific tools like QuickBooks enhances functionality.

You can generate images using ChatGPT Images 2.0 by reverse engineering prompts. Start with an image you like, ask ChatGPT to create a prompt for it. With practice, you improve your prompting skills without needing guides. This method allows for quick and efficient image creation. Works on various images, aside from celebrities.

ChatGPT hallucinations reduced by 52%. Hallucinations occur with hyper-specific queries. High risk in domains like medicine, law, and finance. Current rates down from 20% to about 3%. New model aims to improve accuracy in sensitive topics.

OpenAI's new model emphasizes simpler prompting for better results. Users should adopt outcome-focused prompts over detailed step-by-step requests. Hallucinations in responses have reportedly been reduced by over 50%. The new model 'GPT 5.5 Instant' is set as the default across all plans. Strategies for effective prompting include specifying outcomes and context.

OpenAI Codex can now access and control your computer for automation tasks. It integrates with your browser and can perform complex tasks like analyzing video files. Codex has been updated for a more user-friendly experience targeting non-technical users.
An hour-by-hour walkthrough of how a PM or marketer can use AI through a normal workday. The biggest unlock isn't a single tool - it's a rolling context document the host re-uses across every session. Concrete prompts and screen recordings for each step, including the exact system prompts. A weekly 'audit' habit that keeps the routine from drifting into AI-generated mush over time.