Top 5 AI Tools I Use Every Day as a Product Engineer

There are a thousand "top AI tools" lists and most of them are affiliate-link soup. This isn't that. These are the five tools that have survived months in my actual daily workflow as a product engineer — the ones I'd miss immediately if they vanished. For each I'll tell you what I use it for and where it falls short, because nothing here is magic.

Tools and names move fast, so treat this as a snapshot of how I work more than a permanent ranking.

1. Claude (the model) for thinking, writing, and reasoning

The single tool I open most. Not for code completion — for thinking. Untangling a gnarly architecture decision, drafting a spec, turning a messy brain-dump into a clear plan, reviewing a design for holes.

What I use it for: architecture sparring, writing and editing, summarizing long docs, rubber-ducking bugs, turning vague requirements into a concrete task list.

Why it earned a spot: it's strong at long, multi-step reasoning and follows detailed instructions well, which is exactly what real work needs. I default to a mid-tier model for most of this and escalate to the heavyweight only for genuinely hard problems — the same tiering discipline I described for production agents.

The honest trade-off: it's confident even when wrong. For anything factual or critical, I verify. It's a thinking partner, not an oracle.

2. An agentic coding tool for actual implementation

Distinct from "the model for thinking" — this is the tool that does the engineering work: reads the repo, edits files, runs tests, iterates. I run an agentic coding tool for multi-file features and repo-wide changes, and an editor-native assistant for in-the-moment flow.

What I use it for: implementing scoped features, repo-wide refactors, onboarding to unfamiliar codebases, fixing bugs that span many files.

Why it earned a spot: it changes the unit of work from "lines" to "tasks." I describe the outcome; it does the legwork; I review.

The honest trade-off: garbage spec, garbage output. The skill is writing a precise task and reviewing every diff. I went deep on the options in my post on the best AI IDEs in 2026.

3. Whisper-class transcription / voice tooling

I work on voice AI, so speech-to-text and text-to-speech are both a research interest and a daily utility. Beyond the products I build, I use transcription constantly — turning meetings, voice notes, and rough spoken ideas into text I can act on.

What I use it for: transcribing meetings and voice memos, dictating first drafts faster than I type, and prototyping voice interfaces.

Why it earned a spot: the quality-per-dollar on modern speech models is extraordinary, and dictation is genuinely faster than typing for first drafts. (Latency in these pipelines is literally what my ICANN 2026 paper is about.)

The honest trade-off: accuracy drops on heavy accents, crosstalk, and domain jargon. Always proof transcripts of anything important.

4. An AI image / media tool for visuals

I'm an engineer, not a designer, so for OG images, diagrams, mockups, and quick visual assets I lean on AI image and media generation. It closes the gap between "I need a decent visual" and "I don't want to spend an afternoon in design software."

What I use it for: OG/social images, hero graphics, quick mockups, and — increasingly — short programmatic video. I went deeper on my full media stack, including Remotion for React-based video, in an earlier post.

Why it earned a spot: it removes a whole category of "I'll do it later" friction. Good-enough visuals, instantly.

The honest trade-off: brand consistency is hard, and the output needs a human eye. It gets you 80% of the way; the last 20% is taste.

5. AI-powered search / research

The way I research has changed completely. Instead of ten browser tabs, I lean on AI-powered search and research tools that synthesize across sources and cite them — then I verify the citations.

What I use it for: scoping unfamiliar topics, comparing libraries, finding primary sources, and getting a fast lay-of-the-land before going deep.

Why it earned a spot: it compresses the "what even is this space" phase from hours to minutes.

The honest trade-off: hallucinated or misattributed sources are real. I treat every claim as a lead to verify, not a fact to cite. The discipline is "trust, then verify" — emphasis on verify.

The pattern across all five

Notice what these have in common: each one removes friction from a specific recurring task, and none of them replaces judgment. The model thinks with me, the agent codes for me under review, transcription and media kill grunt work, and AI search accelerates the boring part of research. In every case I stay in the loop on quality.

That's the real lesson. The engineers getting the most out of AI in 2026 aren't the ones chasing every new tool — they're the ones who've found a handful that fit their actual workflow and gotten genuinely good at using them. Depth beats breadth.

What didn't make the list (and why)

I left a lot off on purpose, and the omissions are as informative as the picks:

Single-purpose "AI wrapper" apps. Most are a thin prompt over a model I already have direct access to. If I can do the job in the model directly, the wrapper is just friction and another subscription.
Tools I tried and dropped. Plenty looked great in a demo and didn't survive a week of real use — too slow, too unreliable, or solving a problem I didn't actually have. Adoption is cheap; retention is the real test, and most tools fail it.
Anything that wants to own my whole workflow. I'm wary of tools that demand I do everything their way. The five above slot into how I already work rather than forcing a migration.

The filter, in one line: does it remove real, recurring friction and keep earning its place months later? Almost nothing passes both bars, which is exactly why the list is short.

The takeaway

Five tools, one principle: adopt for fit, not for hype, and never outsource your judgment. If you're building your own stack, start by naming your most repetitive friction points and find the one tool that kills each — then get good at it before adding the next.

For the bigger picture on where all this is heading, see the state of AI in 2026.

Related reading