Every marketing agency owner is sleeping on what Google just added to the Docs API. Your agents can now edit a Google Doc the way a teammate does. They can suggest changes, leave comments, and take action on your comments. Your team accepts or rejects each change. Before this, AI could only overwrite your doc or hand you a copy to compare. A week after Google launched programmatic comments and suggestions, Anthropic put Claude in a Docs sidebar. Put those 2 together and your team can collaborate with agents the same way they collaborate with people. No change management, no training, no new tools. We had built a content agent for an agency client. This week we added Google Docs as a user interface for it. The team types one slash command in the Claude sidebar, then: > Claude loops in the agent via MCP > Its work shows up as suggestions in the doc > Adds comments explaining each change, with the approved source cited > Open questions get tagged for the copywriter, the strategist, or the client Most AI rollouts fail at adoption. Meet your team where they've spent their entire careers working. Adapt to their workflow.

I'm seeing more founders step out of the VC-backed casino to build something that compounds. AI-native agencies and consulting firms can be incredible businesses if you stick with them for long enough. That's what I'm building. Think I can get to $1M revenue/year per employee with a software like profit margin. My only regret is not doing it sooner because of the compounding effect. Building skillsets, AI agents, and relationships focused on one thing creates more and more value over the time. When you wind down a failed startup, or if your VCs take your company from you and install their frat brother as CEO, you have to start over from scratch. When you self-fund, you can build the business and life you want. I don't even want to scale my AI-native agency. I want to make a lot of money but also have time to coach my son's baseball team after school.

"Rip and Replace" looks sexier every day. I only recommend build when you can’t buy what you need. But many SaaS/AI tools have SCAM pricing, so there's probably ROI in 1:1 cloning now because of all the new good and low cost data/dev tools and LLMs. Have a client paying $1,500/mo for a tool that’s buggy and doesn’t do what they need. With the cost savings, plus the benefits of it doing what they need (and actually working), there’s a good case for build because there’s a growth story too. I wouldn’t Rip and Replace to save $100/mo. I also wouldn't have a nontechnical person vibe code it because then you end up in maintenance hell. But getting AI-native engineers to build something that drives revenue growth, not just cost savings compared to the scam, seems a lot more practical lately. Here's how we'll replace our client's $1,500/mo subscription after we get the green light: > Have them record Looms while they use the product > Say what they like and don't like > Run discovery interviews and process mapping to understand the problem they need to solve and the outcomes they want to achieve > Dump all that context into a coding agent to make a Product Requirements Document (PRD) > Show client prototype with example output and get sign off > Build. The objection I hear about builds is “but maintenance”. Valid, but if you if done in an AI-native way and by actual engineers, it shouldn’t be too big of a lift. More importantly, as long as the build is pointed at a valuable use case, it should be worth it

These days, you can pretty much solve any task by defining an eval and hillclimbing over it, instead of directly defining the deterministic/agentic workflow to solve it. The data provider companies' entire job is to define evals for all economic activity to make the frontier models capable of doing anything. Your job then becomes pointing frontier intelligence in the right direction - defining what to solve, and how to measure what good looks like. Agent application interfaces will evolve to capture this. The most complex processes will still need some sort of explicit workflow builder interface, but most tasks can be compressed into goals and eval instructions.

Grok Bot now reads X natively for free (no API spend), so I made it my X content analytics agent. (copy paste the prompt below to make your own) It pulls every post I publish, analyzes and tags the topic, hook, writing style, and format. Then it compares performance to my median. Every Monday morning it tells me: - What's working and why - What's not and why - What to write about next I never open an analytics dashboard. Now I can do more of what works (without going on a scavenger hunt). It's already found some surprising insights that I wouldn't have noticed myself. Next I'm connecting it to site traffic from Vercel Analytics and pipeline from HubSpot, so it can tell me which posts turn into calls and clients. Copy-paste this prompt to Grok Bot to set get it up for yourself: Set up a weekly routine: every Monday morning, review my X posts from the past 7 days (use the last 60 days on the first run to set a baseline). For each original post and quote post, pull impressions, likes, reposts, quotes, replies, and bookmarks. Tag each post's topic, format (list, how-to, story, hot take, clip, quote post), hook, length, and media. Check whether it quoted a viral post or rode a topic that was trending on X that day. Score every post against my median, weighting impressions, bookmarks, reposts, and replies above likes. Then tell me: my top 3 posts and exactly why they worked, what flopped and why, the patterns that keep winning, and 5 post ideas with hooks written in my voice. Keep a running spreadsheet so the patterns sharpen each week. Read-only: never post, like, or follow.

I’ve been in tech for 15 years. I’ve seen so many people rage quit and leave the industry. Startups beat you down if you don’t hit a grand slam soon enough. Here are the common journeys: As an exec/employee, if you don’t have a unicorn on your resume within your first 5-7 years, you become unemployable. As a founder, you often either: - have 3+ failed startups (or base hits), or - you stick at one business for 5+ years before it yields real fruit Regardless of your path, you experience a ton of failure and rejection because it’s simply part of the game: - clients churn - prospects ghost - MRR slopes down - your friends and family lose confidence in you - some people go into debt It's mildly traumatic when you experience this over and over again, without much money to show for all of the 70+ hour work weeks. Only the cockroaches survive 10+ years of chewing this glass. The key to survival is more spiritual than strategic. Many cockroaches become jaded. A shell of their former selves. You can’t let the voices of failure and rejection demons live rent free in your mind. You have to take negative thoughts captive. I promise you, you’re not cursed. Keep going.

Self-Verification is holding agents back from being as good at marketing as they are at coding. Coding agents can objectively prove to themselves that the code works. But marketing is more subjective. You can use an LLM judge but it’s only as good as the human expertise that you teach it.

X is now native in Grok Bot, so your agents can read X for free without paying for API credits. Here are 7 use case ideas: 1. Get a morning brief of the posts taking off in your industry, so you can jump into the convos 2. Pull everyone who liked, reposted, or quoted a competitor's post and filter for your ideal customer profile 3. Do the same with your own posts to see which buyers are ready for outreach 4. Catch buying signals as they happen, like competitor reviews, shopping indicators ("what's the best..."), and complaints about the problem you solve 5. Track every mention of your brand, your clients, and your competitors 6. Get to know a prospect before a sales call by summarizing their last 50 posts 7. Reverse engineer an account's best posts by pulling their top performers and looking for patterns Tell Grok Bot to create "Routines" so these run in the background without you needing to ask.

Built an agent that hill climbs YouTube thumbnails. You can't publish slop and wait for analytics to tell you it was slop. Review has to happen before you publish. But a human can't review every iteration either. That wouldn't be very agentic. So spend the human effort on teaching a judge agent up front. Calibrate the evals once, and the agent can hill-climb without you watching every step. Here's how it works: > nano-banana-pro on fal generates a thumbnail for a HubSpot-style video > A Gemini 2.5 Flash judge scores it on a 5-criterion rubric > The agent reads the judge's notes, writes the next edit, and creates another > Over 4 iterations the score climbs from 4.8 to 8.8 Then a human reviews it. Their feedback becomes 5 rule cards in the judge: > The human-encoded judge re-scores the 8.8 thumbnail to 1.2 > The agent climbs again on its own to 8.4, then 10 > The human approves An LLM judge is only as good as the expertise you give it. The judge didn't need more compute. It needed taste. Calibrate your judge up front, then let your agent hill climb. Self-verifiability is what makes coding agents so good. Marketing is more subjective though, so you need different verification methods.

There are 10,000 AI agents. Nobody knows which ones they actually need. Build vs buy is just the first question. Run your use cases through this 3 step decision tree: 1. Is it generic or unique to your company? - Generic: a standard function every company does the same way. Support inbox, meeting notes, invoices. Buy a vertical agent. - Unique: it runs on your expertise and playbooks. The stuff no vendor has. Build, or configure a horizontal agent. 2. One-off or recurring? - One-off: analytics deep dive, some landing pages, a research report. Buy and configure a personal agent. No need to build from scratch with an agent framework or SDK. Grok Bot is getting crazy good. Integrations are easy. I spun up a half decent marketing analytics agent in an hour. The missing piece is evals. But if it's just for one project, and you're the human in the loop, you're fine. - Recurring: it runs every week, every deal, every client. Build it. 3. How much does a human touch it? The less human in the loop, and the more complex the work, the more engineering you need. Human babysits everything, internal, low stakes, handles basic tasks? Nontechnical person can vibe code it. E.x.: marketing dashboard, lead enrichment script, content repurposing workflow. Little or no human, your playbook and expertise at scale, customer facing, handling a large body of complex work? You need AI engineers. E.x.: healthcare customer support agent, background agents that work while you sleep, custom web apps non-technical teams use all day.

This is the most grotesque attack on a founder I've ever seen. Guy worked at OpenAI, created shareholder value, then had the most successful AI launch of the year. Now AI is trying to destroy him. OpenAI looks less missionary and more mercenary every day.

Everyone says "fix your data before you do AI." Data is only the first of 3 levels of context your agents need. The third, "Sauce", is what adds the most value. 1. Data Systems of record and other raw data sources that give your agents an ontology to understand your business. Ex: CRM, marketing analytics, quickbooks, product usage 2. Context Your company brain that know's the current state of everything. Ex: Call recordings, emails, Slack, positioning docs, customer profiles, etc. 3. Sauce Maps of how decisions and deliverables get made. Ex: Expertise, playbooks, taste. Most Sauce is stuck in your founder or best peoples' heads. You have to get it out. Sauce takes you from AI handling grunt work with a human micromanaging it, to AI driving outcomes without you being the meat proxy between the real world and AI.

AI is massively underemployed right now. Some of the smartest people in the world think AI could end civilization. Yet most companies are just using it to summarize meetings. That's hiring a PhD as an Intern. We need to stop infantilizing AI.

I write 90% of my content by hand. You should too. But if you insist on using AI for content, here's how to un-slop: > Get agent to analyze the writing styles of everything you've published (and/or writing you like) > Create 10+ profiles for common styles, with instructions on when to use each one (store in VOICE .md) > Provide original insights, stories, opinions, etc. (store in POV .md) > Run cron to continuously add and update above as you publish more content (self-improvement loop) > Draft 3 versions of every piece with 3 different style profiles > Edit

Box CEO Aaron Levie (@levie) calls out the MASSIVE opportunity for AI deployment services: "every single one of those companies, whether that's a 50-person firm or a multi-100,000 person firm, is gonna need an army of people to go in and help them with that transformation" "when you go to that law firm and you go to that pharma company and you go to that bank, they need something that bridges the core technology to their workflow in their business process" "somebody has to go into that organization and get it set up, and somebody has to go and provide domain expertise to this model so it really understands our particular business process" Vendor FDEs are incentivized to get you hooked on their platform and to spend more money. Indie FDEs are incentivized to use the best tool for the job and to save you money. Hire indie FDEs.

Super simple way to find use cases for Muse, Grok Bot, or any of the new general agents: 1. Connect it to context rich tools like email and call recordings 2. Give it a brain dump about yourself and your work 3. Tell it to ask you questions about your goals and to fill in any context gaps 4. Tell it to review all of the above and then come up with use case ideas 5. Give feedback, say which you like and dislike and why, get more ideas

Jev is a new kind of AI model. Unlike an LLM, it doesn't generate text, it makes decisions and classifies things. You give it a situation and a set of questions, it answers with a pick and a confidence score. No essay and less room for hallucination. That's what makes it lightning fast and absurdly cheap. $0.042 per million input tokens and output is free. I wanted to take advantage of its near real time speed, so I built a live sales call analyzer. It dissects every word of the conversation, finds key deal signals, then takes action to help the human or AI seller win. Here's how it works: > One Jev evaluate() call continuously runs 18 questions in parallel (TypeSafe calls this "speculative fan-out") > It classifies objections, competitor mentions, qualification signals, and more > High-confidence outputs trigger actions, like grabbing scripts and battle cards, or getting a human sales engineer to jump on the call Jev is free on Vercel AI Gateway this weekend. Happy building.

We built over 50 AI agents and tools for clients. Clients were happy and we were making money, but we shut it all down anyways. We stopped building individual systems. We started building AI Factories for each client. An AI Factory is the system that builds the systems. > The input is client AI use cases prioritized by impact > The output is custom agents and tools (I'm oversimplifying, but that's the gist of it) But now we can't scale any further unless we shut it all down again. We have to stop building AI Factories for each client. We need to build the AI Factory-Factory. Meta, I know, but here me out: The AI Factory-Factory builds a factory for each client. > The input is client problems, growth levers, and business goals > The output is a custom AI Factory that can handle every opportunity (I'm oversimplifying again, but you get it) Now we can deliver absurd value at unprecedented speed, even by AI-native standards. We're kinda disrupting ourselves though, so next we need to start pricing for outcomes because we've completely decoupled time and value.