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.
Gemini 2.5 Pro felt especially strong for deep research and everyday tasks. It seemed better at doing the difficult parts of the work instead of only giving surface-level help. Current Gemini options do not feel as useful to this firsthand user, even if benchmarks say otherwise. The main point is that real daily usefulness can matter more than public performance scores.
The supposed Anthropic “new mode tomorrow” announcement is a parody, not a real product reveal. Claude could be used to recreate the video because the needed model remained available on the Max plan. An earlier recreation of Claude Code’s old “Agent View” mode took about three rounds of instructions. This version required more back-and-forth because Claude struggled to produce workable jokes and a coherent story at the same time, and its first attempt needed major revision.
Gemini 2.5 Pro is still useful for people who like its writing style, especially for turning lecture material into notes. Some people pay for an AI Pro subscription mainly to use Gemini 2.5 Pro in AI Studio. Over the last two days, AI Studio has often returned an internal error when using this model. The quota also appears to run out after only a few requests. The problem has been raised in Google’s developer forum, but there is no clear sign of a fix yet. Gemini 2.5 Pro is expected to be deprecated on October 16, so affected users want the model to work reliably before it goes away.
A person with post-COVID dysautonomia needed to do 4:6 breathing exercises as part of physical therapy. 4:6 breathing means the exhale lasts longer than the inhale. The available online guide videos were not ideal because ads interrupted the session. With no coding skill, the person described the needed flow and used Claude to create a personal follow-along breathing exercise tool. The useful part is simple: a non-developer turned a clear personal need into a small custom aid by explaining it to an AI tool.
Gemini marked a normal question as inappropriate when the question only asked whether Rankin/Bass Productions worked with Warner Brothers. The question was about a company relationship, not adult content, violence, private data, or another sensitive topic. Refreshing the message did not remove the error. This looks like a safety filter mistake, where Gemini treated a harmless question as unsafe.
A firsthand test found Gemini’s image editing results hard to use. The same images and prompts were tried with both Flash and Pro, while the account had a Pro subscription, but the results were still unsatisfactory. The test used the official Mac app. The original conversation was in Portuguese, and the comparison images were translated into English with Nano Banana Pro in Photoshop. With the same images and prompts, GPT completed the edit on the first try, and Nano Banana 2 worked even better through the API.
Gemini displayed error 1097 during use. There are no details about the action that triggered it, the full error message, or the device and app environment. No cause or confirmed fix is available, so the error’s meaning cannot be determined from the given information.
In firsthand use, Gemini had previously produced the best results across many tasks, but its overall performance recently seemed to decline suddenly. The change appeared to begin after the new Flash model launched. No specific failed tasks or comparison results were provided, and there is no evidence that the model launch caused the decline.
A free form builder is being offered for AI agents to create and manage forms. People still get a clean-looking form to fill out, while the agent can read the answers afterward. It does not require a login, and answers can be claimed later through a magic link. It is hosted, so the user does not need to set up their own server. It can be used by pointing an agent to the tool or by connecting through MCP. The intended workflow is for an agent to look at data, create several forms from that data, and then keep track of the replies. The tool is still looking for feedback on what to add or improve.
A creative professional has moved from ChatGPT to Gemini and now to Claude Pro, but feels unsure where to begin. The goal is not to vibe code apps, but to use Claude as a practical assistant for Adobe CC work, daily tasks, and personal organization. Normal chat may not be enough, so better ways to set up Claude as a stronger personal assistant are the main question. The Claude macOS desktop app was only recently discovered, after using Claude through the browser or the terminal. Claude Code has been the only Claude tool used so far, and Cowork may be worth trying if it serves a different purpose from Code.
ChatGPT had previously created long DOCX files and long written outputs without trouble. The same kind of work now returns a message saying it cannot produce even about 2,000 words in one response. The workflow was to analyze a PDF for accuracy, then produce a detailed analysis as either a DOCX file or plain text in the chat. The issue appears in both file creation and normal chat text, not just one format. The open question is whether this is a new limit, a temporary problem, or a request setup issue.
Grok is being discussed as having a very large share of NSFW use, possibly around half of its traffic. Several commenters questioned whether the figure was really half or closer to one fifth. The linked source appeared to be behind a paywall, so the exact method and what counted as NSFW were not clear from the discussion. Some people connected the claim to Grok reducing free image generation and putting more of it behind payment. Others argued that Grok still has strict limits, especially around images and videos, so the actual level of allowed adult content is disputed. The wider point is that consumer AI tools may attract a lot of adult chat, image, and roleplay demand, not only productivity use.
A beginner can know only a few simple prompt habits, such as giving ChatGPT a role or making the request more specific, and still not understand what AI can actually do in professional work. People in fields like social science may want to learn the basics because their work often involves reading, organizing information, analysis, and writing. Finding free courses or tutorials can be hard because many sites look unreliable or scam-like. There is a clear need for trustworthy beginner learning materials that explain how to use ChatGPT in a work setting.
An OpenAI tool no longer appears to let saved library files be added directly. Instead of reusing a file that was already saved in the library, the same file may need to be uploaded manually each time. It is not clear whether this is a removed feature, a temporary interface change, or an issue affecting only some accounts.
There is a need for a place to discuss making video games with AI and Claude Code. The desired community would let people with little coding skill ask experienced developers for help without being mocked or dismissed.
Easy Assign is a freelance platform where students and new workers can find small gigs or post tasks when they need help. The service is deployed on Vercel. In its first 3 days after launch, it reached about 500 users and got some paid tasks. The UI was also edited by hand, but it still looks vibecoded. The main problem is how to make a fast-built AI-assisted product feel more polished and trustworthy.
Gemini Live can be used with a phone camera and other features. However, the steps for finding, turning on, and starting the camera feature remain unclear. No device type, app version, account requirements, or setup instructions are provided, so there is not enough information to use the feature from this item alone.
Claude now shows two separate choices: ‘Opus 4.8’ and ‘Opus 4.8 (1M context)’. The change appeared overnight for at least one person. The practical question is which option to choose. No details are given about speed, price, quality, limits, or whether the larger context version behaves differently in real use.
In a small firsthand comparison, ChatGPT failed to find a pattern. Claude appears to have found it. The available details are very limited, so the exact task, difficulty, and test conditions are unclear. Still, it is a practical example of how two AI tools can behave differently on pattern-finding tasks.
Gemini stopped accepting new image uploads after repeated use with many images. The account had already created more than 150 conversations containing uploaded images. The total number of uploaded images was over 1,200. The cause was not confirmed, so it could be an upload limit, an account usage limit, or a temporary service problem.
Gemini on the web used to work in an incognito tab without signing in to a Google Account as recently as one or two weeks ago. Now, starting a chat in that state can fail after the first reply. The page refreshes automatically as soon as the first answer finishes, so the conversation cannot continue normally. The problem appears tied to using Gemini without a signed-in Google Account, but it is not clear whether this is a policy change or a bug.
An Expo and React Native starter project is presented for building AI chat apps that connect to ChatGPT or Claude. The interface uses dark mode and focuses on chat bubbles that react smoothly to the screen and message content. It also includes “Pro Upgrade” paywall screens for turning free users into paying users. A neon purple accent is used to give the app an AI-style look. The full source code is listed for sale on a separate shop page.
Sonnet 5 refused or stopped when asked to interpret abstract marks on paper or an art image. The same kind of task worked through Haiku 4.5 on the arc.animalabs.ai portal, with no technical setup or system prompt needed. Fable 5 also worked well on a paid account for creative writing, choosing a skill.md suited to the task and producing polished text. This is a firsthand experience about how different models handled the same broad creative request.
Gemini was criticized for taking a set of claims and strengthening them instead of testing or challenging them. The core concern is that an AI tool can make weak or wrong ideas feel more convincing when it simply agrees with the user. For people using Gemini in coding, writing, or planning work, the practical lesson is that a confident answer still needs separate fact-checking.
Claude.ai Research shows a surprisingly small usage difference when the same prompt is used with different settings. Sonnet on low effort often ends at about 49% of the 5-hour usage window. Opus on max effort ends at about 53%. The visible gap is only 4 percentage points. The open question is whether the Research feature itself uses so much overhead that it hides the model difference, or whether the usage display does not clearly reflect the real cost difference.
A firsthand case shows a new Gmail account getting stuck while trying to create a Gemini API key in Google AI Studio. The billing tier appears as unavailable, so the setup cannot move forward normally. The available details do not include a confirmed fix, error screenshot, region, payment setup status, or whether waiting solved it.
Claude’s new TV commercial, “Keep Thinking,” aired during a World Cup game. The ad appears to take a balanced stance on AI, acknowledging both support for the technology and concerns around it. Instead of only promising easy benefits, it seems aimed at making viewers think about how AI should fit into daily life and work.
The experiment asks Gemini to create a completely harmless image that stays as far as possible from its safety limits. Instead of testing how close the model can get to restricted material, it pushes in the opposite direction by maximizing safety. The idea reverses a similar ChatGPT experiment. The available information does not include Gemini’s image or response, so there is no result to assess or compare with another model.
A recent Gemini for Home update appears to have broken smart device control in at least one setup. In Germany, commands such as turning off a light now fail with a message saying Gemini has no access to the devices. The Google Home connection is already enabled in Gemini settings. The problem seems to happen after setup is complete, when Gemini tries to carry out the device command.
Bard is having a problem loading chats. The available information does not include the error message, the cause, a fix, or how many people are affected. The only clear point is that access to chat history or chat screens is not working normally for at least this case.