AI productivity tools: where the time savings actually show up
Search for AI productivity tools and you get lists ranked by feature count. What is harder to find is an honest answer to the question underneath: how much time does any of this actually save?
The measured answer is smaller than the marketing, and smaller than it feels while you work. That gap is worth understanding before you add another tool.
What the measured numbers look like
The best public series on this is the Generative AI Adoption Tracker built by economists Alexander Bick, Adam Blandin and David Deming, published through the St. Louis Fed.
By the second quarter of 2026, 39.2% of employed US adults had used generative AI for work in the past week. AI assisted 6.3% of all work hours. The share of work hours actually saved was 2.2%.
That last number is the one to sit with. Adoption is broad and rising. The measured time saved, across the whole workforce, is roughly an hour a week.
These are self-reported savings, so if anything they lean generous.
Why the gains feel bigger than they are
The sharpest evidence on the perception gap comes from METR's randomized controlled trial of experienced open-source developers.
Sixteen developers worked 246 real tasks in repositories they knew well. Each task was randomly assigned to allow or forbid AI tools. With AI, they took 19% longer. They had forecast a 24% speedup, and after the trial they still believed AI had made them 20% faster.
The researchers are careful about scope, and so should we be: senior developers, mature codebases they already knew deeply. Greenfield work and unfamiliar code may behave very differently.
But the measurement problem generalizes. The parts of AI work that feel fast are visible, and the parts that cost you are not. Waiting for generation feels like progress. Reading output to decide whether it is correct feels like reviewing, not like work you would not otherwise have done.
If you have never measured, your estimate of your own gains is a guess with a hopeful bias baked in.
The categories that hold up
Writing and summarizing. The most reliable category. Drafts, replies, documentation, meeting notes. The savings are front-loaded at the blank page, and the judgment stays with you.
Coding. The highest ceiling and the widest variance. Agentic tools can read a task, implement, run tests and iterate. The METR result suggests the gain depends heavily on context: how well you know the code, how well specified the task is, and how much review the output needs.
Meeting capture. Transcription and action items are solid. They work on input you already have, and they don't reduce the number of meetings.
Research and learning. The weakest case. Output arrives fast and still needs verification, which is work you would not have had before.
Notice what these have in common. AI is good at producing a first version of something you can evaluate quickly. It is worse wherever evaluating the output costs as much as producing it.
The cost nobody measures: handoffs
Here is where the time actually goes for most solo developers and founders, and no tool list covers it.
You write a task in your to-do app. You open the agent. You paste the description into the context. You watch it work. You go back to the to-do app and mark it done. Then you do it again.
Each step is small. None of them is what you would call work. Run that loop a dozen times a day and it is a meaningful share of your week, spent entirely on moving information between two tools that don't talk to each other.
This is also the part of the workflow where measurement is easiest. You can count the copy-paste steps. You cannot easily count whether the generated function saved you 20 minutes.
Closing the loop between your list and your agent
Dovilo is a task manager with focus sessions and a city you build from finished work. The relevant part here is what Dovilo Desktop (Windows and macOS, part of Pro) does for agents.
It runs a local MCP server that gives your coding agent its own task list. The agent lists tasks, reads the project brief, sets status, marks plan steps, writes notes and reports token cost, all from inside the session that is writing code. Claude Code, Claude Desktop, Cursor, Windsurf, Codex CLI and Gemini CLI have presets; any other MCP client connects with a JSON config. Plain MCP, no lock-in.
Three constraints keep it honest:
- Nine tools, read and update only. No create tool, no delete tool. The agent can move work forward, not invent or destroy it.
- Your personal list is not on the tool surface. My Tasks is not filtered out, it is absent. The agent is scoped to one project and its brief.
- It stops instead of guessing. When a decision has consequences, the agent sets the task to
awaiting-user, writes a note explaining the trade-off, and waits.
MCP traffic stays local over stdio, with no Dovilo endpoint in the agent's request path. The AI tasks themselves sync to your phone through Pro cloud sync, which is how you watch steps and token cost while the work happens at your desk. AI tasks run on a monthly quota, with top-up packs from $0.99 that never expire.
Signed webhooks can push status changes into Slack, Linear, n8n or any HTTPS receiver, so the loop closes without you checking anything. Setup takes about 5 minutes: the MCP setup guide. Integration details: dovilo.app/ai.
What Dovilo doesn't do yet
No data export, and no remote MCP, so the agent has to run on the same machine as Desktop. Both are on the public roadmap. If you run agents in CI or on a server, that is a blocker today.
The mobile apps are free on iOS and Android, with unlimited tasks, categories and focus sessions, offline-first storage, five languages and no ads on any tier. Pro is $2.49/month, $19.99/year or $49.99 lifetime, with a 7-day trial started in the app.
How to get real savings
Measure one week before you change anything. Not a feeling, a count. How many hours went to the task, and how many went to moving information around.
Start with your highest-volume repetitive work, not the most interesting. The interesting task is usually the rare one.
Attack the handoffs first. They are the part you can automate with certainty. Whether the model's output was good is a judgment call; whether you had to copy it by hand is not.
Check the output, not the speed. The METR result exists because the developers never measured. Fast and wrong is slower than slow and right, and it does not feel that way in the moment.
Expect single digits, not double. If you are seeing 40%, either you found an unusually good fit or you are estimating rather than measuring. Both are worth knowing.
Frequently asked questions
How much time does AI actually save at work? Across US workers in Q2 2026, self-reported savings came to 2.2% of total work hours, with 6.3% of hours AI-assisted, per the Bick, Blandin and Deming tracker published by the St. Louis Fed. Individual results vary far more than that average suggests.
Do AI coding tools make developers faster? Not always. METR's randomized trial found experienced developers took 19% longer on tasks in codebases they knew well, while believing they had been 20% faster. The researchers caution against generalizing to unfamiliar code or greenfield projects.
Why do people overestimate their AI productivity gains? Because the cost is invisible. Generation feels like progress, and reviewing output feels like reviewing rather than extra work. Without measurement, the estimate is a guess with a hopeful bias.
What actually saves the most time? For most solo builders, removing handoffs between tools. Those are countable and automatable, unlike the quality of a given model output.
How does an MCP server connect an agent to a task manager? It exposes a set of tools the agent can call directly. Dovilo's local MCP server, a Pro desktop feature, gives agents read and update access to a task list that is separate from your personal one, so tasks move without copy-paste.
Can I try this without paying? Dovilo is free on iOS and Android with unlimited tasks, categories and focus sessions. The desktop app and the MCP server are part of Pro, which has a 7-day free trial started in the mobile app.
The tools that save the most time are the ones that remove a step you were doing by hand, not the ones that generate the most. Measure a week, pick your highest-volume work, and connect the tools you already have before adding another.
Dovilo is free on the App Store and Google Play.