Cursor 2.5. Google I/O. Microsoft Build. PwC + KPMG signing six-figure AI seat counts. Plus four boring tools that quietly outperform.
Hey,
This is the busiest week in AI tools of the year so far. Three major dev conferences in one cycle, every default coding model swapped, Google cut AI Ultra by $50/mo, and the Big Four audit firms started signing AI deals measured in the hundreds of thousands of seats.
Five sections + a news rundown — skim what you need.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Are you running your business on incomplete numbers?
Most business owners have financials. Few have financial clarity. BELAY's outsourced accounting team manages your books, tracks key metrics, and delivers timely reporting so you always know where your business stands — and what to do next.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🗞️ WHAT SHIPPED THIS WEEK
Cursor Composer 2.5 dropped (May 18). New model, standard pricing $0.50/$2.50 per Mtok, fast tier at $3/$15. First-week 2× promo runs through May 25 — same window Blackstone × Google announced their $5B TPU joint venture ("N1"). If you're on Cursor, evaluate Composer 2.5 against the promo math now before the second-week pricing kicks in.
GitHub Copilot swapped its default model (May 17). Business and Enterprise accounts now run on GPT-5.3-Codex by default ($1.75 input / $14 output per Mtok, cached input $0.175). If you haven't checked your Copilot bill in 30 days, the per-request cost just shifted under you.
Google I/O 2026 (May 19). Gemini 3.5 Flash and Gemini Omni Flash shipped. AI Ultra repriced from $250 → $200/mo. New $100/mo tier at 5× Pro limits + 20TB storage. Plus $25 pay-as-you-go credit packs. Antigravity 2.0 went GA same day. Read: Google's racing the price floor down.
Microsoft Build 2026 (this week). Microsoft IQ is now GA across GitHub Copilot, Foundry, and Copilot Studios — the context layer that feeds AI agents your workplace data (Work IQ + Fabric IQ + Web IQ). Computer-use agents went GA in 185+ countries on Power Platform, running on Claude Sonnet 4.5 (beta). Frontier Tuning is in private preview.
The enterprise volume tells the story. In the last 30 days: Anthropic + Blackstone $1.5B JV (May 4), OpenAI Deployment Co $4B (May 11), SAP + Claude "Joule" (May 12), PwC + Anthropic for 30K seats (May 14), KPMG + Anthropic for 276K seats (May 19), EY × Microsoft $1B direct (May 21). The Big Four audit-firm matrix is now complete. AI just finished moving from "pilot programs" to "your professional services firm runs on it."
The pattern under all of it: capability is no longer the differentiator — price-portability is. Pick tools that survive a 2-3× pricing reshuffle without breaking your workflow. The tools below all clear that bar.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🔥 TOOL OF THE WEEK — BlackBox AI
While Cursor and GitHub Copilot dominated the headlines this week, BlackBox AI is the boring survivor doing the same job for 20-60% less. It's the AI coding assistant that ships native in 35+ IDEs (VS Code, JetBrains, Android Studio, the works), and the only one routing across 300+ models — Claude Sonnet 4.5, Claude Opus, GPT-5.2, Codex, Gemini, Grok — under a single subscription.
Three things to know:
The price math is silly in your favor. Pro is $10/mo and includes $10 of model credits routing across Claude/GPT/Gemini/Grok, plus Voice Agent and Screen Share Agent. Pro Plus at $20/mo (the popular one) adds $20 of model credits, Multi-Agent Execution, App Builder, the Coding Agent for 35+ IDEs/Web/Terminal, Slack integration, and auto-refill. Pro Max at $40/mo gives you $200 of credits, unlimited agents, SAML SSO, and team analytics. Cursor's equivalent tier is $20-40/mo with one model. Copilot's is $19/mo with one model.
Multi-model routing is the actual differentiator. Most coding tools lock you to one foundation model. With every default model swapping this week (Copilot to GPT-5.3-Codex, Cursor to Composer 2.5), single-model lock-in is a real risk. BlackBox lets you route per task — Claude for refactors, Codex for boilerplate, Gemini for long-context analysis. When one provider raises prices or degrades, you re-route.
Honest caveats. Code quality is inconsistent on complex multi-file tasks vs. Cursor or Claude Code on best behavior. Customer support response times can lag. Data retention controls and training opt-out are reserved for Enterprise. If you're shipping mission-critical production code, treat BlackBox as a supplementary tool, not your primary.
What it's NOT good for: deep agentic coding workflows where you need one model deeply integrated. Claude Code and Cursor still own that. BlackBox is the right pick when you want flexibility + price + IDE breadth.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
⚡ THREE QUICK HITS
Browse AI — Microsoft Build's "agents that touch the web" theme is real, but most of us don't need to rebuild Salesforce agents. We need to scrape a competitor's pricing page weekly and dump it in a Google Sheet. Browse AI is the boring version of that — no-code web scraping and monitoring bots, point-and-click setup, runs on a schedule, sends you alerts when something changes. Free tier covers basic monitoring; paid plans scale by credits. It's been in market since 2020 and is profitable. Every "AI agent that browses the web" startup is reinventing what this already does.
→ Try Browse AI free
Hume AI — voice AI without the conversational-AI hype. Hume's edge is emotion-aware voice — it reads tone, pauses, frustration signals, and adjusts response delivery. Different category from ElevenLabs (synthesis) or Murf (production voiceover). The use case that's actually working: customer support voice bots that can detect when a caller is getting agitated and escalate, healthcare intake calls that read emotional context, sales discovery calls that mark tone-shifts in transcripts. If voice agents are on your roadmap and you've been frustrated by flat-affect responses, this is the differentiated bet.
→ Try Hume AI
AdCreative.ai — Google's UCP Buy Button hit the main SERP this month, which means agentic commerce is now real and ad creative volume is about to spike. AdCreative generates platform-ready banners, headlines, and ad copy from your brand + a few inputs. Starter is $39/mo monthly, drops to $20-29/mo on annual (50% yearly discount). 10 credits/month, 1 brand, unlimited generations, $900/yr iStock library included. Honest flag: video creative is locked behind the $249/mo Professional tier — 6× jump if you need video. Cancel reminders go to spam often, so set a calendar alert if you start the trial. For static ads at small scale, the math works.
→ Try AdCreative.ai free
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
With Big Four firms signing six-figure-seat AI deals, the rest of us are running outbound at smaller scale — but the cold email category is having its own model-shift moment. AI-written sequences, AI-classified replies, AI-driven SDRs. Two affiliates dominate. They optimize for opposite things.
Instantly — per-workspace pricing, unlimited inboxes on every paid plan. Growth at $30-37/mo covers ~5,000 emails/month and 1,000 active contacts. Hypergrowth at $77-97/mo jumps to 25,000 emails. The unlimited-inbox model is the differentiator: a 5-person agency runs ~$77/mo total. Strong warmup network (~1M+ accounts), built-in B2B database, native inbox rotation. Best for: high-volume cold email, agencies, anyone running multiple domains.
Reply.io — per-user pricing, mature multichannel sequences. Email-only plan starts at $49/mo per user for 1,000 emails. Multichannel tiers go higher; Agency plan starts at $166/mo. Jason AI is a more sophisticated AI SDR than Instantly's writer — handles email + LinkedIn + calls + SMS in coordinated sequences, classifies reply intent, books meetings. Best for: SDR teams running account-based outbound with LinkedIn and phone as live channels alongside email.
The honest take: a 5-person team pays ~$77/mo on Instantly vs ~$445/mo on Reply.io for similar volume. That's $4,400/year. If your motion is cold-email-at-volume, Instantly wins on math. If your motion is multichannel ABM with real LinkedIn and call steps coordinated by AI, Reply.io's Jason earns the premium. Most teams overestimate how much real multichannel they do — start with Instantly, upgrade to Reply.io only if you find yourself manually duplicating sequences across LinkedIn.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
✍️ PROMPT OF THE WEEK — The Tier Audit
Google cut AI Ultra $50/mo this week. GitHub Copilot swapped models under everyone's bill. Cursor introduced a new pricing curve. Your AI stack is being repriced in real time — most of you haven't done the math.
Paste this into Claude or ChatGPT with your current subscriptions:
You're a cost-optimization analyst. Here's my current AI tool stack: [LIST: tool, current plan, current monthly cost, primary use case].
For each tool, tell me:
1. Recent pricing changes (last 60 days). Did this provider raise prices, change model defaults, introduce new tiers, or run promos?
2. What I'd pay today if I were signing up fresh. Including any active first-month or annual discounts.
3. The cheaper tier that does 80% of what I use. Specifically — what features would I lose by downgrading, and how often do I actually use them?
4. The next tool category I should consolidate. Am I paying for 3 tools that all do the same job because each one was best at a different moment in 2024-25?
Finally, total my annual savings if I downgraded everything you flagged as overpriced for my actual usage.
Why it works: AI subscriptions decay differently than SaaS subscriptions. The provider changes models, retiers plans, and introduces cheaper tiers behind your back. The thing you signed up for 12 months ago is usually no longer the best value — even from the same vendor. This prompt makes you re-evaluate the deal you're actually getting today, not the one you signed.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📖 FROM THE TOOLCHASE BLOG
Three pieces this week worth a longer read:
Our running log of every meaningful tool update, model release, pricing change, and policy shift. Updated weekly. The single page to bookmark if you want to track this without subscribing to ten newsletters.
With Cursor, Copilot, Claude Code, and BlackBox all moving on the same axis this month, this is the head-to-head we keep refreshing. Includes the price-per-task math nobody publishes.
Our flagship listicle. No fake review counts, no aggregated star averages, no affiliate-paid placements. Just the 30 tools we'd resubscribe to if our credit cards got cancelled tomorrow.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
That's it for this Wednesday.
If you ran the Tier Audit on your stack, hit reply and tell me what surprised you. If you switched coding assistants this week — also tell me. I read every reply.
Talk next week,
Emre
Founder, ToolChase
P.S. Building an AI tool? Submit it for a free editorial review →
P.P.S. Forwarded this email? Subscribe here →