Improve your skills with Growth Memo’s expert weekly insights. Sign up for free!
Am I still a consultant? Or a builder? I have the time available.
My career has changed forever in a way that I am still trying to understand. In the last six months, agetic coding has crossed the threshold. Since then, I have used AI to increase my impact exponentially.
- I designed end-to-end landing pages for a large travel brand that went into production.
- I perform automated keyword optimization, SEO testing, and SEO reporting for my clients with full-featured applications.
- I built an array of useful apps, from automating the SSI (SEO Site Index found in the bimonthly Growth Intelligence Briefs) to Openclaw agents that help me with research and charts.
The work I was sending got better, while it was harder to explain. But if construction costs fall thanks to AI, judgment is the only thing that doesn’t stress. In the meantime, many operators are still hiring, budgeting, and measuring as if execution is mandatory.
I’m not alone: AI companies are reaching $100 million ARR faster than ever, in large part because they’re AI natives. Their entire philosophy of product development is fundamentally different. Heck, Anthropic went from $9 to $30 billion USD in six months and is now worth about as much as Starbucks, Mastercard, or McDonald’s.
And I have my feelings about Matt Schumer’s “Something Big Happening” article, but with a reported 80 million views, it clearly flopped.
So, I want to take a break from publishing research this week and take stock of how agency coding is changing software, distribution, and people.
Impact on Software
In 2024, I made a bold prediction that AI agents will hit 100 million users by 2025. I was out for about a year. Agents did not hit 100 million users in 2025, but came into production in 2026, and the benefits are measurable:
- METR achieved 1.5 to 13x (!) time savings when technical staff used Claude Code.
- 40% cost reduction and 60% time reduction from agent AI is not impossible.
- Bain & Co estimates a 30-50% gain in productivity by outsourcing AI agents and automation.
What happens to software when non-developers can’t post code?
After the iShares software ETF (IGV) dropped 24% in Q1 2026 (the sharpest quarterly decline since Q4 2008), you may feel fear in the air that AI will make software companies obsolete. But software is more than code.
Business software has strong safeguards against AI attacks. Anyone who has ever purchased a CRM or moved to another vendor knows how difficult this is and how much is involved.
Enterprise software is more than code. Code and integration, security, time, sales, and support … all wrapped up in cycles of procurement, IT review, and legal sign-off.
AI can cut into any of those pieces. For example, an agent can manage integration, perform security checks, and even book a demo. But no agent comes forward to be blamed when a mission’s most critical system goes down at 3 AM Accountability is an integral part. Enterprise companies do not change this stack; they build their agents and AI workflows on top of it.
Self-service software is a different animal. Anyone can now browse a simple task tracker in Kanban format. I personally would rather pay a few dollars a month and save myself the trouble of fixing bugs, but it is possible and fast. Do-it-yourself products need to go up in the market. The playbook comes from Notion’s, Figma’s, and Canva’s entry into the business.
In this transition, two archetypes stand out:
- Data providers.
- System of records.
1. Data providers provide value by creating data that the market would otherwise not have access to. These companies lose power from the user interface but gain from their data. For example, let’s say a data provider gives you app store rankings. The user interface of that company is slowly turning into a conflict as many people can’t code their dashboards. But their data becomes more interesting. The strict standards for APIs/MCPs in this world are data integrity, uniqueness, stability, and cost. The logical move is to switch to a headless experience for early adopters and maintain the user interface for legacy users.
2. Systems of record (SOR) is the official place where company data resides. Salesforce, Workday, or Coupa are life problems for many people, but they are billion dollar companies because they are very difficult to change. The moat is a tangle of permissions, test methods, integrations, compliance postures, and decades of workflows built into that data. An agent can produce a CRM in an afternoon; replacing Salesforce in the Fortune 500 is a multi-year change management project. These companies have started and will continue to use AI to provide a better user experience. But their levers are the depth of integration, compliance and research orientation, switching costs, and the quality of their agents. The winners in the SOR space are those whose agents make the existing system of record more useful, not those who try to replace it.
Impact on Broadcasting
Distribution is more important than product, or so the saying goes, but getting it in 2026 is difficult. Platforms are closing (by reducing clicks and keeping users in), and taking opportunities to convert or build direct relationships with off-site visitors.
- AI overview and AI mode make clicks unnecessary and keep users on the Google platform.
- AI chatbots send a small portion of traffic outbound.
- Socializing is fluid, word of mouth is out of control, and payroll is expensive.
From Brand Tax:
Cost per visit increased by 9.4% in 2025 alone, adding up to a cumulative increase of 30% over 3 years. Conversion rates decreased by 5.1%.
How to get distribution in this first world of AI? Two levers are combined:
- Speed.
- Product.
1. Speed means you are doing faster (and better) than your competitors. When all channels of distribution are reduced, and no alternatives are open, the only way to grow is to use them better. Play the game better than the competition. Fast posting speeds become table stakes, and views + counts become differentiators.
PwC found AI’s speed of content generation increases by 3-10x. Simply put, we need to automate more. But not at the cost of trust. If you lose trust, you lose the game.
2. Product is marketing now, with two different results:
- AI sees through the lens of marketing. Agents can read ingredient lists, analyze reviews, and compare specs. “We’re the best X in the world” doesn’t ring true for a screening agent. But solid products are always preferred.
- A free product is the new top of the funnel. Independent tools that solve a real problem are easier to build than ever, and they get better than ads. Ramp Sheets guide users to Ramp’s core product without a marketing budget.
If the product is marketing, the emphasis shifts to product growth: onboarding, engagement, retention. The fastest growing brands these days all have a product-led growth movement. Therefore, marketing and product development melt together.
Impact on People
The power of AI is advancing, but human understanding … is not. Until we get to AGI (God knows when; I hope it’s not anytime soon), human understanding is what hinders the productivity of AI. We can only post as many as we can review.
AI tools can take more input than ever, while our human attention spans are shrinking: AI context windows have grown 3,906x (!) in the last 10 years, from 512 million tokens to 2, while human attention spans have shrunk. We out think faster than we learn to evaluate them.
The two cost curves run parallel to each other: Cost of Automating (radiative decay) vs. Cost of Confirmation (biological bottleneck). In “Some Simple Economics of AI,” Catalini et al., argue that tasks with verifiable outputs will be automated much faster. Work that requires a human to test is slow to compile, so we’ll automate work that’s easy to scale quickly. I feel whenever I use four terminals at the same time: the fixed draw is as high as the output. At scale, what limits us is how much we can test and control.
If anyone can build anything, the way we are measured changes: Skill and tools are less important. But judgment, ideas, and time determine whether you run in the right direction or in circles. It’s very easy to get distracted with AI because the cost of construction is now so low.
Judgment is the non-pressing part. I would ask Claude Cowork to review the contract, but I need to know if it was missed. Claude will happily write the Q4 plan for me, but it’s only good for me to learn which market to attack and what my competitors are going to do.
Over the past six months, I’ve used more automated agent systems than manual work. My clients now have access to unique software they can’t find anywhere else that solves their unique problems.
Three things are now pushing to zero: the cost of building the software, the cost of producing the content, the cost of spinning the tool. But another cost is trending very far from zero: The cost of knowing whether any of them are correct.
I no longer directly “do” work in a normal way. Now I build something that does the job, and I test it. The important work now is the part I can give to the agent … I know what to build, what to kill, and what the agent missed. And I’m here to find out what that means – and you.
Additional resources:
Featured Image: Fit Zstudio/Shutterstock; Paulo Bobita/Search Engine Journal
