The SaaSCity blog
Guides, insights, and resources for SaaS builders. Learn how to launch, grow, and sell your software business.
Build with AI
The Best Vibe Coding Tools for Rapid Startup Prototyping (2026): Pick by Job, Not by Ranking
There is no single best vibe coding tool. There is a best tool for the next 48 hours of your startup. Lovable, Bolt.new, Replit, v0, Base44, Cursor and Claude Code compared by the job they do, what they cost in October 2026, and when the prototype has to stop being vibe-coded.
Build with AI
Gemini 4 Argon: Everything We Know About Google's New Frontier Model (2026)
Google announced Gemini 4 Argon on September 30, 2026, introducing a 1M output token window, a 77.9% state of the art on DeepSWE v1.1, and an introductory API price of $2 per million input tokens and $10 per million output tokens. Access is currently restricted to vetted cyber defenders through the Fairwind Program, with paid API and Ultra access scheduled next. Here is the full breakdown of benchmarks, real-world Google production data, and rollout timelines.
Build with AI
Has Claude Opus 5.5 Been Nerfed? Inside the First Benchmark Built to Prove It (2026)
Six days after Claude Opus 5.5 shipped, an open-source harness called livenerf reached the Hacker News front page attempting to answer the question founders argue about every cycle: does a frontier model quietly degrade after launch? Here is what six days of daily determinism runs actually reveal about model drift, token shrinkage, and vendor risk. If you run a SaaS company on third-party model weights, you are building on a moving foundation, which is why owning your distribution on SaaSCity matters more than trusting a model card.
Build with AI
Claude Code's AGENTS.md Telemetry Gate: What Broke and What Fixed It (2026)
Claude Code v2.1.277 added AGENTS.md support, but tied reading the local file to a remote Statsig telemetry flag. Setups with telemetry disabled silently skipped project rules with zero warnings until v2.1.281 fixed the gate on September 24, 2026. Here is what happened, the complete environment matrix, and the explicit import pattern that avoids remote gates entirely.
Build with AI
AWS Gave the Harness Away. DigitalOcean Started Renting It. (2026)
The same AI model finished 89 benchmark tasks for $56.29 in AWS Strands and $248.05 in Claude Code. The wrapper around your model decides your gross margin, not the model weights. Here is what happened when AWS open-sourced Strands and DigitalOcean launched metered microVM runtimes in the same 48 hours.
Build with AI
Google's AX Just Topped Hacker News: The Agent Runtime That Decides Your AI Bill (2026)
Google's open-source Agent Executor (AX) climbed to number one on Hacker News as developers realized agent SaaS margins depend on runtime scheduling rather than model weights. Here is how AX suspends idle agents, pairs with Agent Substrate on Kubernetes, and reshapes your infrastructure bill.
Build with AI
Spotify Cut Claude Code Token Costs 90% With a Two-Model Router (2026)
A Spotify engineering post went semi-viral this week for a claim that sounds too clean: route the boring parts of a Claude Code session to a cheap model and cut token spend by 90%. It's not a trick, it's a config change, and the same team was refreshingly upfront about what it breaks. Here's the mechanism, the honest failure modes, and how to steal the idea without Spotify's internal platform, plus where a directory like SaaSCity fits into a week this news-heavy.
Build with AI
Don't Ship Naked: The Pre-Launch Security Checklist for Vibe-Coded Startups
AI writes code that runs. It does not write code that resists an attacker. Here is the exact gate to run before your vibe-coded app meets real users — tiered into launch blockers, launch-day requirements, and the stuff that can wait.
Build with AI
Spotify Launched Xirp: One Place to Run Claude Code, Codex and Gemini CLI at the Same Time
Spotify open-beta'd the internal tool 1,300 of its engineers use to run coding agents side by side — every session in its own git worktree, vendor-neutral by design. Here's what it actually does, what it doesn't, and which parts a solo builder can copy for free.
Build with AI
Better Models, Worse Tools: Why Newer AI Models Are Breaking Your Agent Tool Calls
Armin Ronacher discovered that Anthropic newest models are worse at following tool call schemas than their predecessors. Here is why, and what SaaS builders shipping AI agents need to do about it.
Build with AI
AWS Lambda MicroVMs: Why Your AI Coding Tool Needs Stateful Sandboxes
AWS Lambda MicroVMs give SaaS builders VM-grade isolation with container-speed startup — finally a real answer for running untrusted, AI-generated code per user session.
Build with AI
Advanced Claude Code Tips for 2026: Context Hygiene, Plan Mode, Subagents & Power-User Moves
The pro layer of Claude Code — the context-management discipline, plan-mode and subagent patterns, hooks and skills, and the genuinely lesser-known interactive features (rewind menu, side questions, background-subagent control) that separate people who ship from people who fight the tool.
Build with AI
Anthropic's launch-your-agent: Claude Code Agent Deployment From Zero to Live in One Session
Anthropic published a Claude Code skill that guides founders through a complete pipeline — interview, scope, deploy, grade, iterate, schedule — ending with a live Claude Managed Agent in your own Console account. Here's exactly what it does and why it matters.
Build with AI
Claude Code on Large & Existing Codebases in 2026: CLAUDE.md Layering, Semantic Search & Subagents That Scale
The playbook for pointing Claude Code at a big, existing, or legacy repo without burning tokens or losing the plot — hierarchical CLAUDE.md, monorepo exclusions, semantic-search MCP for million-line codebases, Explore subagents, and an onboarding workflow that actually works.
Build with AI
Claude Code Pricing in 2026: What It Actually Costs (Pro, Max, Team & API)
A straight answer on what Claude Code costs in 2026 — every plan from Free to Max to Team, how the usage limits actually work, the API token rates, and which option is cheapest for the way you code.
Build with AI
Claude Code Skills & Plugins in 2026: How They Work, Installing Safely & Which to Use
The practical guide to Claude Code skills and plugins: how skills, plugins, and the two official marketplaces actually work, the exact commands to install them safely, the ones worth using, and how to build your own.
Build with AI
Claude Code Subagents, Background Agents & Agent Teams in 2026: The Real Multi-Agent Guide
How multi-agent work actually works in Claude Code — the three real tiers (subagents, background agents, agent teams), how to create a custom subagent in .claude/agents/, the built-in Explore/Plan/general-purpose agents, cost-routing to cheaper models, and when to reach for which.
Build with AI
CLAUDE.md Templates & Examples That Actually Work in 2026 (Copy-Paste Starter, Rules & Auto Memory)
A real CLAUDE.md reference — what belongs in it, the official 200-line rule, path-scoped rules, imports, auto memory, and a copy-paste starter template you can drop into any project today. Built on the official docs, not folklore.
Build with AI
AI Design Tools in Practice: Why a Jane Street Designer Now Ships in Claude Code Instead of Figma
Edwin Morris, a designer on Jane Street's options desk, now ships working features directly from Claude Code — his Figma usage has 'fallen off a cliff.' Here's the exact workflow, the honest tradeoffs, and what it means for SaaS builders.
Build with AI
Harness Engineering: How OpenAI Used Codex to Ship 1 Million Lines Without Writing a Single One
OpenAI's three-person team generated 1 million lines of production code and 1,500 merged PRs with zero human-written code — by treating the scaffolding around the agent as the real engineering problem.
Build with AI
Headroom: LLM Token Compression for RAG Chunks, Tool Outputs, and Log Files
RAG pipelines dump raw retrieval results into context. Tool agents feed every log line to the model. Most of those tokens do nothing. Headroom compresses them before they reach the LLM — and shapes what comes back.
Build with AI
59% of AI Agent Tokens Go to Code Review, Not Code Generation — New Research
A new study tracked exactly where tokens go in agentic software engineering. The biggest cost isn't writing code — it's the automated review loop eating 59.4% of every token budget.
Build with AI
10 Open-Source Claude Code GitHub Repos Worth Cloning in 2026 (Frameworks, Skills & Workflows)
The Claude Code repos actually worth cloning in 2026 — the official tool, the agentic frameworks that encode real engineering discipline, the skill libraries, and the context tools. With real, verified star counts and an honest must-clone vs nice-to-have tier list.
Build with AI
App Store Connect 'In-App Purchases and Subscriptions' Section Missing? The Complete Fix for Guideline 2.1 IAP Rejections (2026)
Your first subscription has to ship with the binary — but the section that links them has vanished from your version page, and Apple keeps rejecting you for 'no plans available.' Here's exactly why it happens and the step-by-step fix that actually works.