2026 Producitvity Stack
Hey 👋
Every year or so I like to write about the tools I'm using. Partly because I enjoy reading other people's setups, partly because it forces me to think about whether I'm actually getting value from what I'm paying for.
2026 has been a weird one. If I look at my daily workflow compared to even a year ago, the balance has shifted hard toward AI-assisted everything. I'll get to that. First, the foundation.
The Non-AI Stack
These are the tools that keep my day running before any AI gets involved.
Capacities(Notes & Knowledge)
I moved to Capacities for all my note-taking and knowledge management. It fits the way my brain works better than Notion ever did. Instead of forcing everything into pages and databases, Capacities treats each piece of lnformation as an object. A person, a book, a meeting, an idea. They all link together naturally.
I use it as my second brain for project notes, bookmarked references, daily journals, and anything I want to find again later. It is where I log my daily/weekly notes and link them to other relevant objects.
TickTick (GTD & Task Management)
TickTick handles my Getting Things Done workflow. I've tried most of the task managers out there and TickTick hits the right balance between power and speed. Quick capture is fast, the calendar integration works, and widgets in Desktop and Android make it capturing tasks a real breeze. I don't need anything fancier.
Linear (Projects)
Linear is where my projects like Booker Blitz and Kuro live. It integrates with most of the other tools I use (exception being TickTick 👀) and has been my go-to project management tool for years. In the AI age, a project management tool needs to expose the API via MCP in order to stay competitive, Linear not only does this but it also jus released its own built-in agent that can interact with your workspace and I totally recomend checking out.
Locu(Pomodoro & Time Tracking)
I got Locu's lifetime deal when I started a new job that's heavily into time tracking. I use it to do my 45 mins coding sessions and capture work-related notes as I've moved them away from Capacities. Locu is simple, has a Linear integration and uses alot of keyboard shortcuts, which make it a perfect tool for someone who's always context switching.
Helium Browser(Daily Driver)
This one might surprise you. Helium is a Chromium-based browser built on ungoogled-chromium. It strips out all of Google's telemetry, ships with uBlock Origin baked in, blocks third-party cookies by default, and makes zero web requests on first launch without your consent.
It's open source, fast, and the UI is compact. More screen real estate for actual content. Split view is great for working with documentation on one side and code on the other. All Chrome extensions work out of the box (including MV2 extensions, which Chrome itself is killing off). The !bangs feature is borrowed from DuckDuckGo and works offline, straight from the address bar.
There's no built-in password manager and no cloud sync by design. Helium's philosophy is that your passwords and browsing data shouldn't live inside your browser. Which brings me to my next tool.
Proton Pass(Password Manager)
I use Proton Pass as part of the broader Proton suite. Having passwords separate from the browser is the right approach, and Proton's end-to-end encryption gives me peace of mind. I had a 1Password subscription, but their pricing changes, and the fact that my domain's email is hosted in Proton, it was a no-brainer to switch Password Managers too. In the future, I intend to move most of my dev stuff from Google Drive to Proton as I'm on the unlimited plan, but for now, that only lives in my TickTick backlog 😛
Warp Terminal (Multiplexing & Terminal)
Warp is my terminal. It's Rust-based, GPU-accelerated, and the block-based UI is something I didn't think I'd care about until I started using it. Each command and its output is a discrete, selectable block. You can copy a block, share it, collapse it.
The built-in multiplexing means I don't need tmux anymore. I run multiple panes, multiple tabs, and everything persists between sessions. For a workflow where I often have a dev server, a test watcher, and one or two AI coding agents running in parallel, having a terminal that handles all of that natively is a big deal.
Warp has built-in AI features, but my experience with them is that they either hit the limits quite fast, so I don't use them at all.
GitKraken Desktop (Git Client & Agent Management)
GitKraken 12.0 landed in April 2026 with Agent Mode, which changed how I manage parallel coding agent sessions. Before, running multiple agents meant spinning up separate terminals, creating worktrees manually, installing dependencies, and praying you didn't forget to clean them up afterward. Orphaned worktrees accumulate on disk. Status is invisible unless you're in the right terminal window.
Agent Mode abstracts all of that. You give it a branch name, pick your agent (Claude Code, Copilot CLI, OpenCode, whatever you're using), and click Start. GitKraken creates the worktree, runs your setup commands, launches the agent, and leaves your primary working directory untouched. Every active session shows in one panel with real-time status: running, waiting for input, complete.
The evolution in worktrees is the interesting part. GitKraken used to just display them. Now there's an Agent Sessions View that organizes worktrees around agent activity. New worktrees inherit your view settings from the source repo (hidden refs, collapsed folders, etc.) so you don't re-hide everything each time. One-click cleanup removes the worktree and deletes the branch when you're done.
The commit graph shows every active worktree and exactly where each agent is working. No other tool does this. You see full code history, agent status, and what the agent has committed relative to the rest of the codebase. It's the visibility layer between you and parallel agents.
For developers running multiple agents at scale, this is the difference between manageable and chaos. The tool abstracts CLI fluency entirely so any developer on the team can spin up parallel agent sessions, not just those comfortable with git worktree add.
The AI Tools
I'm currently paying for five different AI subscriptions and each one fills a distinct role. A year ago I would have called that excessive. Now it feels like the minimum.
Claude Pro (The Planner)
Claude is my main thinking partner. I use it through Claude.ai with the Pro subscription, and it handles far more than coding.
Blog posts, planning documents, brainstorming sessions, this article you're reading right now started as a conversation with Claude. When I'm working on Booker Blitz, Claude helps me think through game design decisions, balance mechanics, and structure the JSON modding system. It's the tool I open when I need to think clearly about something complex.
For non-coding work, Claude is unmatched. The writing quality is a step above everything else I've tried, and the ability to give it context about my projects and preferences means I spend less time repeating myself.
Perplexity Pro (Augmented Search)
Perplexity replaced Google for most of my research. When I need to understand something, compare options, or find current information with sources, Perplexity gives me cited answers instead of a page of blue links.
The part that sold me is Comet, Perplexity's AI browser. Comet is Chromium-based with an AI assistant built into the sidebar. The automation capabilities are what sold me. I use it for social media interactions: drafting LinkedIn posts, responding to comments, summarizing articles into shareable formats. You give it a high-level instruction and it handles the multi-step workflow. Need to summarize an article and turn it into a Twitter thread? Comet does it without tab-switching or copy-pasting between tools.
I keep Helium as my daily driver for browsing and Comet specifically for research and social media workflows. Two browsers, two jobs.
GitHub Copilot Pro (The Workhorse)
I've been using Copilot for years at this point. It was one of the first AI coding tools I adopted, and the cost-to-value ratio has been excellent for inline completions and chat.
The big story right now is the pricing change. Starting June 1, 2026, GitHub is moving Copilot from request-based billing to usage-based billing with AI Credits. Your $10/month Pro subscription gets you $10 in AI Credits. Usage is calculated by token consumption across models, so heavier models burn through credits faster. Code completions and next-edit suggestions stay unlimited, but chat, agent mode, code review, and the cloud agent all consume credits.
This is a significant shift. The community reaction has been mixed, with some developers pointing out that $10 worth of API-rate tokens doesn't go far during heavy agentic coding sessions. GitHub paused new sign-ups for Pro and Pro+ while they roll out the new billing infrastructure.
What I use most from Copilot right now is the cloud agent. I kick off a task, switch to something else, and come back to review the result. This has changed how I work in a fundamental way. I spend much more time reviewing code and asking for adjustments than I do writing code from scratch. The dynamic has flipped from "I write, AI assists" to "AI writes, I review and steer." ⚠️ Add specifics about what kinds of tasks you delegate to the cloud agent, or a concrete example of the review workflow.
OpenCode (Access to Cheaper Models)
OpenCode is an open source AI coding agent built in Go. It runs in the terminal as a TUI (built with Bubble Tea, which any Go developer will appreciate) and supports 75+ models across every major provider.
What drew me to it is OpenCode Go, their $10/month subscription that gives you access to curated open coding models. The models have gotten good enough that for many tasks they're close to proprietary alternatives, and the session limits are generous. DeepSeek, Qwen, GLM, all accessible through one subscription with reliable hosting.
If the session limits on OpenCode Go stay this generous, I'll likely cancel my Claude Code subscription. For the kind of agentic coding work where I'm firing off tasks and reviewing results, the quality gap between the best open models and Claude has narrowed enough that the price difference matters.
OpenCode also has LSP integration, persistent sessions that survive terminal disconnects (thanks to the client/server architecture), and vim-style keybindings. It feels like a tool built by terminal-native developers for terminal-native developers.
Gemini Pro (Images & Life Admin)
Google's Gemini handles two things for me that nothing else does well.
First, image generation. When I need a quick visual for a blog post, a social media graphic, or concept art for Booker Blitz, Gemini generates it. The quality has reached a point where it's good enough for most non-professional uses.
Second, life integrations. Gemini connects natively to Gmail, Google Maps, Google Calendar, and the rest of the Google ecosystem. For personal life admin (scheduling, email triage, route planning) it has context that other AI tools can't match because it's sitting on top of all your Google data.
I don't use Gemini for coding or writing. It has its lane, and it stays in it.
The Workflow in Practice
A typical day looks something like this:
I open Locu and start a planning session, then I open Capacities and start filling out the daily note/updating the weekly note, doing this, I send any relevant tasks to TickTick, which I open next and plan my day using time blocks. I pull the tasks I want to work on from Linear into Locu and planning is done.
I open Warp and spin up my dev environment. If I have a feature to build, I'll describe it to the Copilot cloud agent or OpenCode and let it start working. While it does, I switch to Claude to plan the next feature, write documentation, or work on a blog post. Perplexity handles any research I need mid-task. Comet handles social media and browser interactions when I take a break from coding. Gemini handles anything personal that comes up.
The pattern that's emerged is delegation and review. I spend more time describing what I want, reviewing what I get back, and iterating on the output than I do producing the first draft of anything myself. Whether that's code, prose, or design decisions.
This is a different kind of work. It requires clear thinking about what you want before you ask for it. The throughput, though, is higher than anything I could achieve alone.
That's the stack for mid-2026. I'll probably revisit this in six months because the rate things are changing, half of these tools might not even exist by then. Or I'll have added three more subscriptions. Either way, I'll let you know.
If you're using any tools I haven't mentioned, I'd love to hear about them.