7 Ways to Get So Good at AI, People Will Think You Are AI

Sam Liang, CEO of Otter, criticized a journalist for using iPhone Voice Memos and manual transcription instead of his AI-powered service. The article outlines seven strategies for becoming an "AI native" at work, including using advanced agents like Codex, adopting voice input, and sharing extensive data with AI tools to maximize personalization and productivity.

Sam Liang is appalled as I confess my technique for recording an interview: running the Voice Memos app on an iPhone https://www.wired.com/tag/iphone/ and transferring the transcript manually to a Google Doc. The CEO of Otter https://otter.ai/ , a transcription service for analyzing meetings, looks at me as if I tried to call into our video chat using a rotary phone. He believes, naturally, that I should switch to Otter. He’s probably right. It’s all part of a new identity at work and maybe at home https://www.wired.com/story/meet-the-sad-wives-of-ai/ : the AI native. Time-saving productivity tools https://www.wired.com/story/6-ai-based-productivity-tools-tested/ like next-gen note-takers, task-based agents, and chatty inbox assistants are exploding in popularity as they invade every nook and cranny of our digital lives. While it’s critical to keep concerns about security https://www.wired.com/story/ai-workplace-privacy-security/ and hallucinations https://www.wired.com/story/reduce-ai-hallucinations-with-rag/ top of mind when using any AI feature, early adopters are developing a fluency that will likely pay dividends for years to come. Being AI native—or “agentic,” as AI natives say—means staying adaptable to new experiences. Transcription failures aside, I’ve embraced experimentation, from generating AI podcasts https://www.wired.com/story/ai-podcast-google-notebooklm/ to letting Claude organize my desktop files. Some of this I talked about in my newsletter series last year, AI Unlocked https://www.wired.com/story/artificial-intelligence-tools-ai-unlocked-newsletter-season-two/ . If you want to get so good at using AI tools that your coworkers start questioning whether it's blood or ribbon cables running beneath your skin, here are my seven tips for AI-powered ascendance. 1. Kill Your Chatbots ChatGPT https://www.wired.com/tag/chatgpt/ is so 2022. These days, the cool kids are all about Codex https://www.wired.com/story/openai-really-wants-codex-to-shut-up-about-goblins/ . Your eyes may glaze over, rightfully, at the mention of “AI agents,” but compared to anything on the market even a year ago, software automation tools like Codex and Anthropic’s Cowork https://www.wired.com/story/anthropic-claude-cowork-agent/ are leagues better at actually taking over your computer and completing tasks. Don’t waste your time fiddling with a single chatbot when you could be commanding a whole army of them. 2. Go Voice Mode Oh, you’re still typing up everything you want your AI tools to do, Boomer-style? That’s cute. But trust Otter’s Liang: “Voice will become more dominant https://www.wired.com/story/ai-kill-phone-screen/ moving forward,” he tells me. “People hate writing.” He caveats that I , a journalist, probably don’t hate writing https://www.wired.com/story/tech-reporters-using-ai-write-edit-stories/ , which is, mostly, true. This move is primarily for the input, not necessarily the output. I rarely use the voice-only mode on ChatGPT, for instance, but I often speak a prompt into my phone and then skim the written output. 3. Build a Sandbox Even though agents are actually good now, the rascally little devils can still eff everything up without proper boundaries. Earlier this year, a Claude-powered agent deleted a startup’s entire production database https://x.com/lifeof jer/status/2048103471019434248?s=20 and backups. So if you’re ready to have some external entity take control of your computer, you need to spend an afternoon researching everything https://support.claude.com/en/articles/13345190-get-started-with-claude-cowork these tools can do and set up some dedicated folders with the files you want them to access. 4. Give It Everything You’ve Got With apologies to our privacy-minded security writers https://www.wired.com/category/security/ , it’s simply the case that the more data you share with AI, the more personalized the outputs can be. Jo Barrow is the chief of staff at Granola https://www.granola.ai/ , one of Otter’s competitors, and she puts it this way: “I have a personal OS system, which is a series of files on my computer which my AI lives inside. Whenever I ask questions, all of that context is right there, and the agent can go and figure it out. I don’t need to repeat myself over and over again.” Fair warning: Sensitive conversations are still best had without a permanent record. 5. Create an Impersonator Barrow tells me she dumps all of her Slack https://www.wired.com/tag/slack/ messages into a document to let bots know how she sounds on that platform, and she does the same for her email inbox and social media accounts. “People use AI for finessing their tone of voice,” she says. “There’s only so many times you can say, ‘OK, a little bit warmer. OK, a little less formal.’ That’s a big time sink.” Creating these guides for the agents to follow won’t fully replicate your voice, but they can nudge the bot to output something at least closer to your cadence and tone. 6. Think Across Teams Data is powerful, and adding more of it from people around you can further enhance AI tools. Consider your coworkers: “Many people are using a meeting note-taker now, but they're still using it at the individual meeting level,” Liang says. He touts the “knowledge engine” Otter can create when a whole workplace buys in, from the engineering team to the marketing department. You can even do this at home: If family members pour various notes from their day into a single, shared AI tool, that’s going to provide more insights than siloed usage. 7. Learn to Jailbreak Successfully using AI tools in 2026 doesn’t require writing—I mean speaking —perfect prompts. Even so, starting off more complex tasks with a creative, well-calibrated request can be clutch. Experiment with the wording, especially if you’re hitting unexpected guardrails https://www.wired.com/story/deepseeks-ai-jailbreak-prompt-injection-attacks/ blocking the output. Recently, I attempted to convince a bot to send me email addresses for a variety of niche experts, and it refused to deliver. But when I started a new chat and shared details about why I wanted this info for reporting purposes, of course, not stalking , it forked over a list. What Say You? Let us know what you think about this article in the comments below. Alternatively, you can submit a letter to the editor at email protected /cdn-cgi/l/email-protection 573a363e3b17203e2532337934383a .