In partnership with
METR ran a study on experienced developers using an AI assistant. On certain tasks, they came out 19% slower than coding by hand. The tool built to write the code for them made them slower. I'll tell you why at the end.
Half your market is one app away.
Your business is already on Instagram, SMS, and web chat. But 52 million immigrants in the US rely on WhatsApp to connect with businesses they trust — not email, not phone calls.
Wati helps you show up on WhatsApp and every channel they use. Are you still not there?
This week
Code generation is solved. Codex, Claude Code, and the rest produce code that compiles, runs, and clears the easy tests. You can assume the machine can write code now. But writing code was never the same skill as building software. Google's whitepaper this month, "The New SDLC With Vibe Coding" (Addy Osmani et al.), says it flat out: the line between vibe coding and real engineering isn't whether you use AI. It's how much verification, judgment, and structure you wrap around the output.
The paper has a name for where this bites: the 80% problem. An agent rips out roughly 80% of a feature in minutes. The last 20%, the edge cases, the error handling, the integration points, the subtle correctness, needs context the model doesn't have. And the failures changed shape. They're not syntax typos anymore; they're conceptual: wrong assumptions about the business logic, a missed edge case, a structural decision that quietly rots the codebase six months later. The dangerous part is that the code looks right and often passes the basic tests. A fluent wrong answer slips through exactly where a visibly broken one would get caught.
That is the 19%. The experts weren't slower because the AI is bad. They were slower because the job moved: from writing, to reviewing, guiding, and correcting. The paper splits the new role into two modes you switch between all day. Conductor: you're in the editor, directing every change in real time. Orchestrator: you define the goal, hand it to agents, and review what comes back. Both reward the same scarce skills, and none of them is typing speed: specification, decomposition, evaluation, system design. Generation went free. Judgment became the bottleneck, and the bottleneck is what gets paid.
Here's the move for tomorrow morning. Before you write your next agent prompt, write the three conditions that define "done" first. Tests verify the deterministic part (this input produces that output). Evals verify the rest (right trajectory, right tools, right quality bar) through scoring rubrics and an LM judge. Without both layers, you're vibe coding no matter how clever the prompt is. The 20% the agent can't finish is yours. That's where the craft lives, and where the money is.
In the news
Google published "The New SDLC With Vibe Coding" (Addy Osmani, Shubham Saboo, Sokratis Kartakis, May 2026). Past the buzzword, it puts numbers on the shift: field surveys report 25 to 39% productivity gains, but the picture gets honest fast once you measure the time spent verifying. The role it describes isn't implementer anymore. It's conductor and quality arbiter.
This week on the blog
The Part Your AI Agent Can't Handle Is the Business You Haven't Built : where the agent stops, judgment starts. That gap is where the money is.
Wasted AI Credits Are Every Builder's Dumb Tax. Here's What I Do Instead : directing agents without torching the budget. The orchestration part, in practice.
Every Generation Picked the Wrong Stack. Vibe Coders Are No Different : why the stack choice is still a human decision, not an AI output.
Investors see ANOTHER return from Masterworks (!!!!)
That’s 6 sales in 7 months. 29 all time. And the performance?
16.5%, 17.6%, and 17.8%, net annualized returns on sold works held longer than one year (See all 29 at Masterworks.com)
It’s not from stocks, private equity, or real estate… it’s from contemporary and post war art. Crazy, right?
With Masterworks, you don’t need to be a BILLIONAIRE to invest in multi-million dollar art anymore.
Historically, the segment overall has had attractive appreciation and low correlation to stocks.*
Masterworks targets works featuring legends like Banksy, Basquiat, and Picasso, identifying what they believe to have significant long-term appreciation potential, not just at the artist level but at the level of individual artworks.
As one of the largest players in the art market, with $1.3 billion invested over 500 artworks, they pass critical advantages through to their 70,000+ members to add art to their portfolios strategically.
Looking to diversify your investments in 2026?
*According to Masterworks data. Investing involves risk. Past performance is not indicative of future returns. See important Reg A disclosures at masterworks.com/cd.
Revenue Stack: June 2026
What survives a model swap is a clean workflow: a deploy path, a test suite, verifiable success conditions. The model is interchangeable; the verification ramp isn't. It's also what carries you through the 80% problem, the messy last fifth no agent finishes alone. The book's 8-step Blueprint builds exactly that ramp: the discipline that turns raw generation into software that holds.
The machine writes the code. The book teaches the part it doesn't.
Where is your AI workflow on the spectrum?
Phil
PS: reply with what you actually verify before you merge generated code. Tests? Review? Nothing? I'm collecting the most serious setups for a future edition.
Back to that 19%. The experts weren't dragged down by a broken AI. They were doing the new job already (verify, correct, direct) without having built the harness for it yet. Generation went free. The craft just moved to the other side of the screen.
You're receiving this because you signed up at rentierdigital.beehiiv.com.