Welcome back
I have some personal news to share this week. But first, I want to talk about something that keeps coming up in my work — why most companies are failing at the thing everyone's excited about, and what I think it actually takes to get AI working inside a real engineering org.
Let's dig in.
AI News Roundup
↗ Claude Opus 4.6 Ships with Agent Teams
Anthropic's latest release brings 1M token context, agentic coding improvements, and the ability to coordinate agent teams. The context window expansion is particularly significant for the enterprise use cases Jain describes.
↗ Cursor Hits $1B ARR with 2.1M Users
The agentic IDE crossed a billion in revenue, with NVIDIA moving 40,000 engineers onto Cursor workflows. The tooling layer is officially mature, which makes what I'm about to say even more important.
Some personal news…
I'm launching Seeko.
It's a hands-on AI consulting practice where I work directly with companies to figure out where AI creates real leverage in their engineering and operations, and then we build it together. Not slides, not training sessions — actual systems that your team owns when we're done.
I've been doing this work quietly for a few months now, and something keeps coming up in every single engagement that I want to talk about. I wrote the longer version on the Seeko blog, but here's the short take.
The Gap That Nobody's Talking About
Boris Cherny, the creator of Claude Code, revealed his workflow in January — he runs 15 parallel Claude instances and ships over 100 PRs a week. The General Intelligence Company reported engineers shipping 20 PRs per day. Karpathy even rebranded his "vibe coding" concept into something more deliberate: agentic engineering.
The numbers are real, the tools are powerful, and everyone has access to them. But here's what I keep seeing in my work: most teams that adopt these tools can't come close to replicating those results, and the gap has almost nothing to do with which tool they picked. It has everything to do with what they built underneath it.
It's More Boring Than It Sounds
I know that's not what people want to hear, but the teams getting extraordinary results have invested heavily in things that don't make for good tweets — clean documentation, review discipline, architectural clarity, and making sure their agents actually know what they're working on.
The teams struggling are the ones that skipped all of that and went straight to the shiny part.
Here's the thing nobody wants to talk about: the agents writing code is maybe 20% of the actual work. The other 80% is everything around it. Addy Osmani put it well when he said the single biggest differentiator between agentic engineering and vibe coding is testing. That's not a fun conference talk, but in my experience, it's the whole game.
And the most common issue I see when I walk into a company isn't the model or the prompts — it's what the agents actually know about the codebase, the product, and the organization. It's almost always a mess. Gartner predicted that 60% of AI projects would be abandoned due to data that simply wasn't ready for AI, and Arvind Jain's piece this week makes the same argument — without a solid foundation, agents just produce what he calls "work slop." I think
They're both right, and it maps perfectly to what I'm seeing on the ground.
Why This Is Actually Good News
Here's where I get optimistic. If the bottleneck were model quality, you'd just have to wait for the next release and hope for the best. If it were about picking the right tool, that's a gamble.
But the bottleneck is actually engineering fundamentals — documentation, testing, review discipline, architectural clarity — and those are completely learnable. AI-native engineering is more boring than it sounds, and I think that's the point. The companies winning right now aren't the ones with the best tools; they're the ones doing the boring work that makes those tools effective.
Everyone has the camera. The question is whether you'll do the work to learn how to use it.
Work with Seeko
This is exactly what I do at Seeko. I come in, look at how your team works with AI tools, identify where the gaps are in your workflow and infrastructure, and then we build the systems to close them together. No handoffs to a junior team, no mystery deliverables — you own everything when we're done.
I'm keeping the practice intentionally small because this kind of work requires real hands-on attention, not scaled consulting theater. I have capacity for 2 more clients this month, first come first serve.
You're the right fit if you're a CTO, VP of Engineering, or founder at a company with 50-500 employees where your team is using AI tools but not getting the results you expected, and you suspect the problem runs deeper than the tool itself. If that sounds like where you are, tell us what you're working on. The form takes two minutes, and I personally review every submission.
Now I Want to Hear From You...
I'm genuinely curious about this one.
This week's question: What's the biggest bottleneck you're hitting with AI tooling right now? Is it the tools themselves, the infrastructure underneath them, getting the team bought in, or something else entirely?
Hit reply with your thoughts. Best responses get featured next week.
Until next week,
Haroon