We pick and plainly summarize new features, pricing, usage limits, and policy changes across major AI tools — Claude, ChatGPT·Codex, Gemini, and Cursor — from a solo developer and maker’s point of view.
Using voice to text with Claude can make everyday speech look much less clear once it becomes written text. Spoken thoughts may feel normal in the moment, but the converted text can look broken, vague, or poorly organized. The practical lesson is simple: voice input is convenient, but important prompts may need a quick cleanup before sending.
Cursor and VS Code used with Claude can both feel good enough for light coding. The clear downside raised for Cursor is that it uses a lot of memory. The main question is whether Cursor has features that VS Code plus Claude cannot offer, and why some developers choose Cursor as their only coding tool.
A firsthand report says the ChatGPT apps on macOS and iOS do not show the Create Image feature as a visible option. The apps can still make images when the prompt clearly asks for an image. The main difference is the workflow. On the web version, Create Image can be selected directly, and then almost any prompt is treated as an image request without needing to say “create an image.” The feature is available on the web, but not visible in the same way in the Mac and iPhone apps.
The Codex Mac app keeps asking for permission even when the approval policy is set to “Never.” The expected behavior is that Codex should continue without asking for more approvals, but the app still interrupts the workflow. The setup uses only the Codex app, not the CLI. The real question is whether the Mac app can fully disable permission prompts, or whether its permission behavior is separate from that setting.
Claude Sonnet 4.6 on medium effort can handle a large amount of daily work and smaller high-level tasks, based on firsthand use. Bigger models may still make sense for very hard questions or deep reasoning, but routine work can feel cleaner with a less powerful model. The experience comes from someone who does not see themselves as an elite power user, but uses Claude more heavily than most people. In that use case, going above Sonnet has rarely felt necessary.
An old Gemini chat still appears in My Activity, but the answers shown inside the Gemini app are now completely different from what they were before. The issue is not simply that previous chats cannot be found. The same chat is visible, but Gemini’s earlier responses look as if they changed overnight. The user is looking for a way to restore the original content or fix the display problem because the chat is important.
Two screenshots are circulating with a claim that Claude may require passport information for use. The claim is not confirmed in the supplied material. There is no official Anthropic notice, country list, affected account type, start date, or exact requirement included here. The only solid takeaway is that a passport-information requirement is being questioned, but its truth is still unclear.
adtention.ai is an ad network aimed at developers. It puts ads inside places where developers already work, such as the terminal, an IDE, or an AI chat app. The maker built it after normal paid ads did not work well for their own product. The first go-to-market idea was a prompt-to-earn plugin for Claude Code. Several projects are now exploring ways to add adtention.ai to their own products. Each project can decide whether ad money goes to users, becomes extra revenue, or helps fund open source work. Developer-relevant SaaS products can claim 3 free ad blocks.
Claude for Word shows a “connected files” list when several Word documents are being edited. That list lets each document’s chat session be opened. The problem is that working across several documents still requires clicking back and forth between separate Word files. The main question is whether Claude for Word has a cleaner way to coordinate a multi-document or multi-agent session, similar to Claude Code.
The key issue is how Claude Code CLI works with a subscription plan. Based on the available content, the practical question is whether Claude Code CLI is included in an existing Claude subscription, requires separate payment, has usage limits, or needs a specific sign-in setup. No concrete answer, price, setup step, or limit number is included in the provided item.
The top ProductHunt launches on June 16, 2026 centered on support, ad creation, sales, and developer workflow tracking. Zoona AI received 109 votes for automating customer support by learning from documents and past conversations. Tadka received 83 votes for helping teams make more ad creative without hiring a design team. Voice Calls in Chatwoot received 79 votes for putting calls, chats, and emails into one support inbox. ClientJam received 78 votes for AI-powered lead generation for designers and agencies. Glint received 69 votes for showing Claude Code activity in the places where a user wants to see it.
The post's title claims Claude Code has been improving impressively, but it contains no supporting details, examples, or data beyond the title itself.
After a recent update, Bard/Gemini’s safety filters may be behaving differently. There are no concrete examples showing what became more restricted or less restricted. The main point is that safety-related limits seem different from before the update.
Mercor is hiring backend engineers to evaluate frontier AI coding models on real software engineering tasks and backend system design. The role is listed as remote and full-time, with pay described as equivalent to $85 per hour. Each accepted task pays $400, and typical tasks take 2 to 3 hours after ramp-up. The work involves using AI coding agents to solve complex backend engineering tasks and checking model-generated code for correctness, maintainability, and performance. Engineers also need to find bugs, edge cases, and architecture failure modes. Applicants need at least 2 years of professional backend development experience and experience with APIs, microservices, distributed systems, or backend platforms. Regular use of AI coding tools such as Cursor, Claude Code, Codex, Windsurf, or Gemini CLI is required, and experience with large-scale production systems is a strong plus.
Mercor is hiring remote contract machine learning engineers to test and improve advanced AI coding models. The pay is listed as about $85 per hour, or $400 for each accepted task. After ramp-up, a typical task is expected to take 2 to 3 hours. The work involves using AI coding agents to complete and judge realistic machine learning and AI engineering tasks. It includes reviewing code for model training, inference systems, MLOps, and LLM applications, then finding bugs, slow parts, and edge cases. The role also compares results from multiple frontier AI models. Applicants need at least 2 years of professional machine learning engineering experience, experience with production ML systems or AI products, and regular use of tools such as Cursor, Claude Code, Codex, Windsurf, or Gemini CLI. Experience deploying machine learning systems to production is treated as a strong plus.
ChatGPT is being used as a tool for making original characters. The current main tool is ChatGPT, though other AI programs are also in the picture. The creative process is enjoyable, but people nearby either dislike AI or do not care about it. There is interest in finding others who use AI for the same kind of character creation. No specific workflow, prompt, result, or tool comparison is included.
Nano Banana 2 can create a vertical phone-style selfie where the camera sits very low and points upward. The person is placed near the bottom center of the image, while a fuzzy cat or dog leans in from the top of the frame upside down and looks directly toward the lens. The scene uses a bright modern apartment, large windows, soft daylight, and a clean background. The image should keep skin, fur, and phone-camera lighting realistic, with the two faces balanced in the frame. The prompt also blocks common failures: eye-level shots, flat angles, animals shown from the side, cartoon styling, broken body shapes, extra limbs, and scary expressions. To get the pet facing the right way, the wording needs to clearly say that the pet is upside down and looking straight down at the camera. To keep the whole composition working, the camera angle must be described as very low and pointing up at both faces.
The focus is which work setup people use when running Codex. The options include using the plain Codex CLI with no extra layer, using Pi agent, using a desktop app, or using another setup. There is no comparison result or recommendation yet. The value is in seeing which tools real Codex users choose around Codex for day-to-day work.
Mercor is hiring remote contract data engineers to judge how well AI coding agents handle real data engineering work. The pay is listed as $80 per hour equivalent, with $400 for each accepted task. After ramp-up, typical tasks are expected to take 2 to 3 hours. The work involves using AI coding agents to solve complex data engineering tasks and then reviewing the results. The review covers ETL pipelines, data warehouses, analytics platforms, and distributed data systems. The role also requires finding bugs, scalability problems, and failure modes, then comparing answers from several newer AI models. Applicants need at least 2 years of professional data engineering experience, hands-on experience with data pipelines or data platforms, and regular use of tools such as Cursor, Claude Code, Codex, Windsurf, or Gemini CLI.
The central question is how modern researchers use AI tools in real research work, and what the pros and cons are compared with traditional methods. The area of interest is animal science and a possible independent business that builds oral bait machines for feral cats to reduce disease and other health problems in colonies. The request seeks practical experience from animal science researchers or researchers in other fields. It also reflects that AI use in academia is still a debated topic.
Codex Account Manager is a small desktop app for managing several Codex accounts in one place. It is meant for people who use different accounts for different workflows and do not want to keep logging in and out or editing settings by hand. The tool stores separate account profiles and helps switch between them in a more predictable way. It was built as a simple utility to remove one repeated annoyance, not as a large product. The project is open-source, so others can inspect the code or build on it.
Mercor is hiring infrastructure engineers on an hourly contract to use and judge AI coding agents on real cloud and reliability tasks. The role is aimed at DevOps engineers, SREs, platform engineers, and cloud engineers. The work includes using AI coding agents to complete infrastructure tasks, then reviewing AI-made answers involving cloud platforms, Kubernetes, CI/CD, observability, and infrastructure automation. Engineers need to find bugs, reliability problems, missed edge cases, and compare answers from different frontier AI models. The listing is aimed at people who already use tools such as Cursor, Claude Code, Codex, Windsurf, or Gemini CLI in their work. Pay is listed as $85 per hour, or about $400 for each accepted task. Eligible locations include the United States, United Kingdom, Canada, Australia, and many countries across Europe and Central America.
Mercor is hiring iOS engineers for contract work focused on testing AI coding agents. The work involves using AI coding tools on realistic mobile engineering tasks and comparing the results from several advanced AI models. Engineers review Swift code for quality, speed, and ease of maintenance. They also look for bugs, missed edge cases, and weak design choices. The listed pay is equivalent to $85 per hour, with $400 for each accepted task. After ramp-up, a typical task is expected to take 2 to 3 hours. Applicants need at least 2 years of professional iOS experience, production experience with Swift and modern iOS frameworks, and regular use of tools such as Cursor, Claude Code, Codex, Windsurf, or Gemini CLI. Shipping real iOS apps is listed as a plus.
The available content is very thin. It says $800 was spent in connection with Claude, and the spending felt worthwhile. It does not show which plan the money went to, what work it supported, or what time period the cost covered. No concrete results or comparison with other tools are provided.
Building an app with AI help did not make hosting simple. The project started as a side project inspired by GitHub examples and ideas. One possible setup was Vercel for the frontend and Render for the backend, but that looked expensive if the app grew. The chosen path was to host the app through Docker on a rented server. A Contabo VPS in the Australia region cost about 153.41 AUD, or 108.48 USD, for one year. A .com domain from Cloudflare cost about 14.79 AUD, or 10.46 USD, for one year. Dokploy was installed on the VPS, and GitHub was connected so that pushing code to the main branch publishes the app through Dokploy. Each push to main now restarts the whole app as part of the deployment flow.
The latest ChatGPT app can be used to restore an old photo with a few simple steps. Start a new chat, upload the photo, and ask ChatGPT to bring back color, sharpen details, raise the image toward 4K quality, and remove borders or visible damage. The instruction should also say to keep the original look and character of the photograph. The example shows an old photo of a late mother being turned into a clearer, cleaner version.
Some Claude skills are promoted as if they deeply change what Claude can do, but the actual content may be very simple. A skill can sound like an advanced workflow tool, then turn out to be close to a basic instruction such as thinking carefully before answering. Even inside the Claude community, there is a joking frustration about how much hype can surround small prompt-like additions.
The ClaudeAI community is showing fatigue with posts that feel written by AI. The main concern is that polished AI-style writing can remove human voice, small imperfections, emotion, and personality. AI is still treated as a useful tool that helps people do more, but communication can feel less real when everything sounds machine-made. The practical point is that people may need to bring back more direct, human expression even when they use AI to help write.
Google AI Studio users are running into confusion over the current rate limits for Gemini Pro and Flash models. Without a Google AI Pro subscription, it is not clear how much usage is allowed before limits apply. A related experience shows that even someone with the Google AI Pro 5TB subscription can hit a “rate limit” message while using Google AI Studio for writing. The tool’s output was useful for creative writing, but practical information about limits was harder to find than for some other AI tools. The concrete issue is that model limits can interrupt work inside Google AI Studio, and the difference between subscription access and actual AI Studio limits is not obvious to users.
Claude is being used in day-to-day work, especially for client reports and progress updates. The main question is whether Claude has helped people build working websites that are useful to themselves or others. No specific website examples, build steps, or results are included in the item.