I track every tool subscription. Not because I’m cheap—because most tools are garbage dressed up with good marketing.

Here’s what actually earned its keep in 2025.

The Winners

Cursor ($20/month per seat)

The AI-powered IDE that actually works. Our developers ship 30-40% faster. Not hype, measured in PRs merged.

Before Cursor: 3-4 PRs per developer per week After Cursor: 5-6 PRs per developer per week

That’s real productivity. The $240/year pays for itself in the first week.

Linear ($8/user/month)

Still the best issue tracking for engineering teams. Fast, opinionated, gets out of your way.

We tried going back to Jira for “enterprise features.” Lasted two weeks before the team mutinied. Linear stays.

Notion AI ($10/user/month add-on)

Skeptical at first. Now it writes first drafts of documentation, summarizes meeting notes, and helps organize messy pages.

Is it life-changing? No. Does it save 2-3 hours per person per week? Yes. Worth it.

Loom ($15/user/month)

Async video replaced 40% of our meetings. Record a 5-minute video instead of scheduling a 30-minute call. Everyone watches at 1.5x speed.

The math: 40% fewer meetings = 40% more deep work time. Loom is underrated.

Superhuman ($30/month)

Expensive email client. But email is 2+ hours of my day. Superhuman saves 30 minutes daily. That’s 10 hours/month for $30. Easy decision.

Not for everyone. If email isn’t a core part of your work, skip it.

The Disappointments

Jasper AI ($49/month)

AI writing tools have gotten commoditized. Claude and ChatGPT do the same thing for cheaper or free. Jasper’s moat disappeared.

Cancelled after the free trial.

Monday.com ($36/seat/month)

Tried it for cross-team visibility. Too slow, too complex, too expensive. Notion does 80% of what we needed for a third of the price.

Various “AI Agent” platforms

Tested five different AI agent platforms this year. Total spend: $3,000+. Useful output: nearly zero.

The technology isn’t there yet. Save your money for 2026.

Expensive analytics tools

We paid $400/month for an analytics platform. Switched to PostHog’s free tier. Zero regrets.

Most startups don’t need enterprise analytics. You need to know what users do, not 47 different attribution models.

The Surprising Value

GitHub Copilot ($19/month)

Less flashy than Cursor but more stable. For junior developers especially, the autocomplete suggestions are genuinely useful.

We use both Cursor and Copilot. Belt and suspenders approach to developer productivity.

Figma ($15/editor/month)

Not new, but the Dev Mode features that shipped this year made handoff dramatically better. Less back-and-forth between design and engineering.

Claude Pro ($20/month)

Better than ChatGPT for complex analysis and longer documents. Anthropic has nailed the sweet spot between capability and reliability. I use it daily for thinking through problems, drafting strategies, and research.

The API is good too, but the Pro subscription is faster for ad-hoc work.

The Framework

How I evaluate tools now:

1. Does it save time that I can measure? Not “feels faster”—actual hours.

2. What’s the switching cost? Easy to try, easy to leave = low risk.

3. Does the team actually use it? Adoption matters more than features.

4. What’s the total cost at scale? $10/seat seems cheap until you have 50 people.

Tools that fail these tests get cancelled immediately. No sunk cost fallacy.

2026 Predictions

What I’m watching:

• AI coding assistants will keep improving. Budget $50-100/developer/month.
• “Productivity suite” bundling will increase. Microsoft and Google will make it harder to pick best-of-breed.
• More tools will have good-enough AI features built in. Standalone AI tools may struggle.

But honestly? The best tool advice hasn’t changed: pick something, use it consistently, optimize later.

Tool switching costs more than suboptimal tools.