Using AI to Work Faster

AI tools are everywhere right now, and many professionals are either using them poorly, or not at all. Here’s where AI speeds you up, slows you down, and how to use AI for work effectively.

AI chatbots like ChatGPT, Claude, and Gemini are pattern matchers trained on massive amounts of text (e.g. all of Wikipedia.org). When queried, they predict the most likely response based on what they’ve read. They might sound smart, but they don’t know what’s actually true about the real world.

Where AI for Work is Useful

AI works best when the task is generic, the volume is high, and the material already exists.

Generation Tasks

Volume work. Any task where you want lots of variations but would normally settle for only a couple of variations because it’s too laborious. AI removes this constraint and can generate options lightning fast. However, the trick with volume is to ask for more than you need. The more information you give the AI, the better the output will be. If you can explain the inputs in great detail, and explain the output you’d like to see in great detail, the AI will have much less trouble filling in the process to get from A to B.

• “Here’s a product name. Give me 15 alternatives: 5 descriptive, 5 punchy, 5 abstract.”
• “Here are 12 email subject lines. Rewrite them so they all follow the same format: question + benefit. Keep them under 50 characters.”
• “Check this 30-page document for spelling, grammar, and inconsistent capitalization.”
• “Take this a series of 7 emails between me and a vendor. Generate 10 possible responses.”

Reformatting. Any task where you want to reshape existing material into a different format. The information already exists, you’re just changing the presentation. This is one of the best use cases for AI since the info already exists, the AI doesn’t have to guess, it just uses what you give it. Again, be specific with the end-goal, if you’re looking for one sentence, or three bullet points, or two paragraphs, specify that.

• “Turn this 5-page campaign brief into a one paragraph summary for the executive team.”
• “Read this meeting transcript and pull out a succinct list of decisions and action items.”
• “Take this long blog post and turn it into a five-tweet thread.”
• “Reformat these customer interview notes into a few punchy quotes I can use in a landing page.”

Brainstorming. AI is great at providing something to get you started and get you out of analysis-paralysis mode. Ask for more than you need, and you’ll probably get something that starts your brain moving so that you can keep making progress on a task. Granted, it may not get you to the end-goal every time, but very often it can help you start.

• “Give me 30 possible names for a new customer loyalty program.”
• “List 10 angles I could use for a Q4 holiday campaign.”
• “Suggest 15 questions I could ask in a stakeholder interview about this initiative.”

Consumption Tasks

Drafting. First versions of anything you’ll edit heavily. Emails, memos, briefs, outlines, agendas. Faster to edit a draft than stare blankly at an empty page. For reasons we’ll talk about in the next section, I wouldn’t recommend shipping anything written by AI.

• “Draft a follow-up email to a client who hasn’t responded in two weeks. Friendly but direct.”
• “Write a one-page creative brief for a new campaign based on these notes.”
• “Draft an agenda for a 30-minute kickoff meeting covering scope, timeline, and roles.”

Querying. Use AI like a more advanced Google that can also read documents/files you give it. Drop a 200 page report and ask it specific questions. It’ll point you to the exact page you need. Instead, treat the responses as starting points to dig deeper, not as final answers.

• “Here’s a CSV of last quarter’s ad spend. What are the top three channels and how did they perform against goal?”
• “Look at Acme Co’s website. Compare and contrast Acme with its competitors, determine the pros and cons of their positioning.”
• “Search my previous chats and find where we discussed the Q3 email strategy.”

Learning. Treat AI like your personal tutor. It may not be an expert at every field, but it can personalize its lessons to your needs. Faster than reading a bunch of random online articles. It can bring you up to speed quickly so that you can then dive into whatever topics you want more clarity on. This is the same idea behind behavioral economics in marketing, where small shifts in framing change how information lands.

• “Give me a summary of the state of the art of attribution modeling right now, and explain it like I’m five.”
• “I have a meeting about supply chain logistics in an hour. I know nothing. Give me 10 things I need to understand and 5 questions I should be asking.”
• “Walk me through how a Roth IRA works.”
• “Explain what a SOC 2 audit involves like I have no technical background.”

Where AI Falls Short

Strategy. AI averages across all of its training data and gives you the most generic answer. It’s not very good at coming up with new ideas, so asking the AI to do anything that requires genuine creativity doesn’t really work. Admittedly, that bar is rising every day. For today, I’d still recommend using AI to brainstorm and come up with ideas, but not to decide which course of action is best. Frameworks like our crawl, walk, run approach to marketing personalization still depend on human judgment about where your team actually is.

Originality. AI recombines patterns that already exist. Sure, some of it may look novel, but if you look closely, you’ll see all the existing patterns. As a result, AI outputs start to look similar, the writing is very distinctive, and the ideas follow safe patterns.

Accuracy. The technology is getting better, and is sometimes scarily accurate. But it still confidently makes facts up and tries to pass them off as real. Verify all statistics, sources, quotes, names, dates, case studies, etc. This is what people mean when they say AI is “hallucinating”, you could just call it lying. The thing is, the AI doesn’t know when it’s wrong, so it won’t be able to verify itself.

Voice. You may already know the AI voice: safe, generic, sanitized, with liberal use of em-dashes (these things: —). They mimic the rule of three excellently, frustrating writers everywhere. They’ve perfected the corporate voice. They sound good. But a little too good. Like there’s no perspective, no experience, no opinions behind the words. It all starts to sound robotic and boring.

Skills That Matter Now

Asking better questions. The quality of output is determined by the quality of the prompt. Include context (who’s the audience, what’s the goal), constraints (length, tone, what to avoid), and your own thinking (“here’s what I’ve tried, here’s what I’m stuck on”). “Write me an email” gets slop. “I need a follow-up email to a client who hasn’t responded in two weeks, we met once, friendly but not desperate” gets something usable.

Knowing what good looks like. AI will produce output for any prompt. If you can’t tell good from mediocre, you’ll ship mediocre. In fact, the people getting the most out of AI aren’t the best prompters, they’re the best editors.

Sharing examples. If you want output in your voice, give it your voice. For example, paste five things you’ve written and tell it to match the style. The more examples, the less it defaults to bland.

Verifying everything that matters. Every fact, stat, source, quote, name. If it’s going into a deck or in front of a client, verify it. Assume sources are fabricated until you click them.

The AI for Work Tool Landscape

Chatbots. Pure thinkers. Claude, ChatGPT, and Gemini are the current big ones, and they’re all roughly the same for most work. Claude is the best at coding, ChatGPT has the largest ecosystem, Gemini integrates with Google Workspace. Pick one, or better yet, use all three.

SaaS wrappers. Tools that put a chatbot (same as above) behind an interface. Good for specific tasks. The underlying tech is usually a chatbot doing the thinking, with an interface designed to make it easier to use. Worth paying for if the interface saves you real time, otherwise you can get the same functionality from a chatbot.

Agents. The next evolution, e.g. Manus, ChatGPT Agent, Perplexity Labs. This is AI that thinks, but can also act. Browser agents, research agents, coding agents. Still early in development, they’re as unreliable as chatbots from a few years ago, but improving fast. I’d recommend you use chatbots for most things, but begin to test out agents when you can.

Where AI for Work Is Going

The chatbots I mentioned are already developing persistent memory to maintain huge amounts of context. Multimodal input (text + image + video + etc) is already being developed. Real-time data access is advancing quickly. Agents are early in the lifecycle, but can already do complicated tasks.

Honestly, nobody has a great idea of where this is going. AI now accounts for a record share of S&P 500 market cap, the technology receives more investment than anything else, and the technology is literally getting better every day. The big companies (Anthropic, OpenAI, Google) are jostling to take the pie, so pay attention to what they’re doing. New possibilities are being invented each month.

How to Start Using AI for Work

Pick one task you’ve been doing by hand and try it with an AI chatbot this week. Ask it to do the whole thing, and it’ll help you figure out the limitations. Then, compare the result to your usual version. If AI is worse, you’ve learned the boundaries. If AI is faster, you now have a new tool to speed up your work.

As you work with the chatbot, you’ll figure out what works for you. Save prompts you use that you find effective into a document. Continually update that document with other tips and tricks. For example, I have “be as succinct as possible” and “stop agreeing with everything I say, I find disagreement more useful”. Any other tips you can come up with to tailor the responses of the AI can be added to this document.

Then, upload this document at the beginning of any chat, or to the context section of the chatbot, and it’ll respond to you in the way you want.