Topic
5 news articles · 7 tutorials
On June 10, Claude Code v2.1.172 let sub-agents spawn their own sub-agents, five levels deep, and the patches through June 26 show how load-bearing that depth cap already is. I have run seven interactive CLI agents at once on one box for weeks. Recursion is real, and it multiplies the three things that already break a fleet: cold prompt-cache, dropped sockets, and the supervision tax.
GPT-5.5 and Claude Opus 4.7 are peers now. Not equivalent, peers. The math on Indian operator workflows says you want both running, called from one terminal, with a router that picks per-task. This is the field write-up after two days of wiring a Claude-to-Codex bridge, plus the rules I learned the hard way.
Anthropic shipped Fast Mode for Claude Code on Opus 4.7. The press read said faster output. The operator read is sharper: same Opus, same 1M context, output speed roughly 4x. Twenty-one days of pair-programming against a real backlog, here is the operator playbook for when to flip /fast and when to leave it off.
Four AI coding surfaces, four pricing tiers, one question for the Indian solo shop: which two do you actually pay for? We ran every tool through real tasks for thirty days and the answer is not the one Twitter keeps yelling about.
A 2.1.121 bump isn’t supposed to burn Tuesday afternoons. But when the subagent fork flag collides with legacy hooks and plugin loaders, the audit pass becomes the cost nobody budgeted for.
I run a fleet of interactive Claude Code and Codex CLI sessions from a single orchestrator, all in detached tmux panes. This is the exact read-inject-digest loop I use in production, the three Linux blockers that killed my GUI approach before this, and the stuck-pane gotchas I hit running ten sessions at once.
A working operator names the exact AI skills that pay in India in 2026 and the hype that wastes your six months. Honest signal for students, job-seekers, and exam aspirants deciding where to invest effort.
Claude Code is sharp until it hits the edge of your machine and your private tools. I wrote three small MCP servers in Python to close that gap. Here is the real pattern, the real gotcha that bit me, and what it costs.
Fanning out subagents in Claude Code looks free until you hit the cap or your forks clobber each other's commits. These are the real fixes I learned running fanouts: the fork env flag that shares the parent's cache, the WebFetch cache leak, and the worktree pattern for parallel writers.
A non-traditional operator running real production sites on Claude Code, Coolify, PocketBase on Oracle ARM, and Cloudflare. The real workflow is Opus as orchestrator plus subagents executing, governed by hard gates: verify before claiming done, snapshot before destructive ops, and a flat refusal of --no-verify and --force. Here is what works, and the failure modes a non-engineer hits shipping to prod in India.
I installed Claude Code on a stock Ubuntu 24.04 box, set up Pro OAuth, and shipped a real refactor with it. This is the install that worked, including the bits the official docs skip.
I run Claude Code every working day across a dozen-plus projects from one Linux box in India. Here is the real Max 20x economics for a solo operator, the one env flag that fits 2x more agents in the same cap, /fast mode on Opus, and the 5-hour meter that punishes too many headless runs.
Your privacy, your call.
No tracking runs unless you allow Google Analytics and ads. Decline and they stay off. Read our privacy policy.