What I Watched
March 1 – March 7, 2026
Agent orchestration wars, Claude Code's feature velocity, Obsidian memory stacks, and the one-person AI business playbook.
When the Tools Start Fighting Each Other
Every week I export my YouTube watch history and run it through an analysis pipeline to see what my subconscious was actually paying attention to. Not what I planned to learn — what I kept clicking on. This week, the pattern was unusually clear: the AI tooling space is in a moment of genuine competitive tension, and my feed reflected it.
92 videos from March 1st through March 7th. After filtering out the entertainment and automotive content, what remained was a concentrated picture of a specific question: as AI agent capabilities mature, where does the leverage actually live? In the platform (Claude Code), in third-party orchestration layers (OpenClaw), in personal knowledge infrastructure (Obsidian), or in the business model that wraps all of it together?
This week's content was effectively four different answers to the same question. And the fact that I kept returning to all four suggests the answer is probably: all of the above, in sequence.
Claude Code Is in a Feature Velocity Phase
The volume of Claude Code feature content this week was different from previous weeks — it wasn't tutorials on existing functionality, it was coverage of new capabilities shipping in real time. Nate Herk's 'Claude Code 2.0 Is Finally Here' set the tone. Remote Control shipped and immediately generated both enthusiasm ('Claude Code Just Added What Everyone Wanted') and scrutiny ('Claude Code Remote Control is here... But Has A Problem. Here's How To Fix it.'). The /loop and /simplify commands generated their own debate — Chase AI's 'The Claude Code /loop Misinformation is Hilarious' and AI Coding Daily's /simplify tutorial both pointed at a community actively stress-testing new features in real time.
The skills ecosystem matured significantly. DesignCourse's 'Use Claude Code DESIGNER Skill to 10x UI Designs,' Nate Herk's 'Claude Code Skills Just Got Even Better,' and Brock Mesarich's '7 Claude Code Skills I Can't Live Without' all reinforced that skills are becoming the primary leverage point for power users. The BMad Method Masterclass gave a comprehensive IDE workflow framework. And multiple videos addressed context management — 'I Changed One Thing and Doubled My Claude Sessions' and '99% of Claude Code users waste hours rebuilding context' — suggesting this remains the dominant friction point for serious users.
Claude Code Worktrees in 7 Minutes from Developers Digest was a quiet standout — a crisp explainer on a capability that dramatically changes parallel development workflows. The through-line: Claude Code is shipping fast, the community is stress-testing just as fast, and the gap between users who track these updates and those who don't is widening every week.
Claude Code's feature velocity in early March — Remote Control, enhanced skills, contested /loop and /simplify — marks a platform inflection that rewards users who stay current.
The OpenClaw Question: Platform vs. Orchestration Layer
OpenClaw dominated my feed in a way that felt like watching a new product category form in real time. Brian Casel's 'How to create JOBS for OpenClaw agents' appeared three times — I watched it multiple times, which the data correctly flagged. Alongside it: 'OpenClaw vs. Claude for Running an Agent Team,' 'Day 7 Vibe Coding: Building OpenClaw Dashboard,' Nick Vasilescu's pieces on OpenClaw wrappers as a business model in 2026, and GTM AI Academy's case study of replacing $14K of agency work with a Claude setup. The content volume alone signals this is worth understanding deeply.
The counter-narrative appeared just as strongly. 'I Replaced OpenClaw With Claude Code in One Day' and 'Claude Can Now Do OpenClaw Natively (Remote Control + Tasks)' both argued that Claude Code's Remote Control feature renders OpenClaw redundant — or at least compresses its advantage significantly. This is the classic platform vs. layer debate: does the platform eventually absorb the capabilities of the tools built on top of it? For OpenClaw, the answer appears to be arriving sooner than expected.
Both camps are right in different contexts. OpenClaw's job-based agent model has a specific UX and workflow that appeals to non-technical operators. Claude Code's native capabilities appeal to developers who want to stay in their existing environment. The question isn't which is better in the abstract — it's which matches your operating model. That nuance was mostly absent from the content, which is probably why I kept watching.
OpenClaw and Claude Code native capabilities are converging rapidly — the right choice depends on whether you're optimizing for operator UX or developer workflow.
Obsidian Is Becoming the AI Memory Stack
The Obsidian thread from last week didn't just continue — it deepened. 'Claude Code + Obsidian = UNSTOPPABLE' from Chase AI, 'Claude AI Reads My Obsidian Second Brain. I Just Vibe' from Luke Skyward, 'Organized Your Ideas With Claude Code AI Agent and Obsidian' from Tony Huang, and James Goldbach's short on using Obsidian to give Claude Code memory all converged on the same architectural insight: Obsidian isn't the destination for your notes, it's the persistent context layer that makes Claude's outputs personal rather than generic.
Nick Milo's 'DON'T Make These 3 Obsidian Mistakes' and the Tool Use podcast episode on Coding Agents in Obsidian added useful texture — the mistakes framing in particular signals a community mature enough to be doing it wrong in systematic ways. Chase AI's 'Claude Code + NotebookLM = CHEAT CODE' was an interesting variant, suggesting the underlying pattern (structured personal knowledge as AI context) extends well beyond Obsidian specifically.
The phrase that stuck with me from Luke Skyward's video: 'I just vibe.' It's not ironic — it's describing a workflow where the infrastructure does enough of the heavy lifting that the human can operate at the ideation layer rather than the execution layer. That's the end state this entire stack is optimized for. Obsidian, structured correctly and connected to Claude, is one of the clearer paths to getting there.
Obsidian connected to Claude Code isn't a productivity hack — it's the persistent memory architecture that transforms AI from a generic tool into a personalized thinking partner.
The One-Person AI Business Has a Documented Playbook
Dan Koe opened the week with 'How To Start A One-Person Business With AI (From 0 To $1 Million)' and 'A realistic way to earn with AI.' Brock Mesarich followed with 'How I Run a $70K/Month Business With No Team (Claude Code).' Nate Herk added 'How I Build $10,000 Apple-Style Websites with Claude Code' and 'How to Sign AI Workflow Clients (With 0 Followers).' Dan Martell's 'If I Wanted to Make F-You Money This Year, I'd Do This' and Brett Malinowski's 'YC Was Right: AI Can Run an Entire Business Now' completed the cluster. This wasn't coincidence — it was a thesis being argued from multiple angles simultaneously.
What struck me was the specificity. These aren't 'AI will change everything' pieces — they're documented playbooks with revenue numbers, client acquisition strategies, and tool stacks. The Nate B Jones pieces on stopping competition with 400 applicants and building in a weekend reinforced the same point: the one-person AI business is a repeatable architecture, not an outlier story. Sabrina Ramonov's 'You're wasting $10K/year on software AI replaced for free' and 'Software Engineer Jobs Are Rising' added useful counterbalance — the opportunity exists, but so does the noise around it.
The most conceptually interesting piece in this cluster was Solo Swift Crafter's 'Harness Engineering: The Skill That Will Define 2026 for Solo Devs.' The harness engineering framing — building the scaffolding that lets AI execute reliably rather than writing the code yourself — reframes the skill that actually matters in 2026. It's not prompt engineering. It's system design at a level of abstraction most developers haven't shifted to yet. That concept quietly tied together everything else I watched this week.
The one-person AI business isn't aspirational anymore — and 'harness engineering' is the meta-skill that separates those building it from those just talking about it.
Where It All Converges
Four threads that look separate on the surface share a single underlying logic: leverage in AI-native work is shifting from knowing how to use tools to knowing how to architect systems. Claude Code's feature velocity rewards those who build on its evolving platform rather than locking into specific workflows. The OpenClaw question is fundamentally about which orchestration architecture fits your operating model — a design decision, not a tool preference. Obsidian as AI memory is about building infrastructure that compounds, not just taking better notes. And the one-person AI business playbook is, at its core, a systems design problem.
The through-line is harness engineering — building the scaffolding that makes AI systems reliable and compounding rather than using AI for one-off tasks. Whether it's Claude Code worktrees for parallel agent execution, Obsidian vaults as persistent context, or job-based orchestration in OpenClaw, the practitioners generating real results are the ones who've built the harness first and added features second. That's what I kept gravitating toward this week, and it's the lens I'm bringing to everything I build at Acceleration Works.
Notable Videos This Week
What This Means for My Content
The concept that landed hardest this week was harness engineering. Not prompt engineering, not vibe coding, not AI workflows in the generic sense — the specific discipline of building the scaffolding that makes AI systems reliable and compounding. It's the reason Claude Code worktrees matter, it's the reason Obsidian as a memory layer matters, and it's the reason the one-person AI business is a systems design problem more than a tool selection problem.
That's the frame I'm bringing to everything I build at Acceleration Works. The advantage in 2026 isn't which AI tools you use — it's whether you've built the harness that makes those tools work together reliably while you operate at the strategy layer. This week's content was, in aggregate, a field manual for what that harness looks like across different domains.
The competitive edge isn't which AI tools you use. It's whether you've built the harness that makes them reliable.
built the harness that makes them reliable.