A few months ago, buying a new dishwasher meant a browser tab with too many reviews and a rough mental comparison. Now it means an HTML dashboard. Structured, filterable, actually useful. I did not plan for this to happen. It happened because the friction of producing structured output dropped to almost nothing.
That drop changed everything about how I work. Not in a vague, future-of-work way. In a specific, daily, I-do-not-open-my-apps-anymore way. Let me walk through it.
Claude became my interface to everything.
I do not open most of my apps anymore. Word, Apollo, my file system. They still run. They still do their jobs. I just stopped going in. Claude sits in front of all of it. I describe what I need. Claude operates the tools.
Word is the clearest example. I produce a high volume of documents: proposals, scoping briefs, client deliverables. I have not opened Word to draft or edit anything in months. I describe the document, Claude writes it, formats it, saves it. The file exists where it should. I never touched Word.
Apollo works the same way. Every action, from building lists to managing sequences to checking records, happens through Claude. I go into Apollo to verify. I never act there myself.
Once that pattern set in with a few tools, it spread to everything else. Which brings me to one of the more mundane problems it solved.
File chaos became a weekly AI process.
My directories were a sprawl. Months of files with no consistent home. The kind of mess that builds up quietly when you are juggling client work, a new house, and a newborn son, and filing things properly is never the thing that wins the priority call.
Every week now, Claude runs against my folders and organises them, moving files into the right places based on rules I have built up over time. The pile stopped growing.
Finding files follows the same logic. I describe what I remember, what it was for, roughly when I saved it, and Claude finds it. No search boxes. No guessing. No giving up.
That same logic, describe the outcome and let Claude handle the process, eventually spilled over into my personal life too. And that is where the dishwasher comes in.
The dishwasher story: AI research running in the background
We moved into our house just over a year ago. Our son arrived at the same time. A new house and a newborn in the same week. Everything that was not immediately urgent got pushed to the bottom of the list. Buying a dishwasher sat there for twelve months, somewhere below "fix the bedroom door" and "figure out the vaccination schedule." When you are running on three hours of sleep and trying to keep a small human alive, appliance research does not make the cut.
When we finally cleared enough of the backlog to get there, the old version of this process would have looked familiar to anyone who has done it. Two people independently browsing Amazon, sending each other 20 URLs, no shared criteria, no way to compare, and a decision that never quite lands. We had done exactly this with our washing machine a few months earlier. It took two weeks.
This time, I described what we needed (child lock, specific cycle settings, a certain capacity) and asked Claude to browse Amazon and build a comparison dashboard. I kept working. Claude ran in the background, pulled the relevant models, checked the specs, and produced an HTML file with filters for every criteria we cared about.
My wife opened the dashboard. She filtered by child lock. Three options surfaced. We picked one in ten minutes.
The difference is not that we made a better decision. It is that the research happened while I was doing something else.
That experience changed what I bother to make. Before, building a structured comparison for a personal decision felt like overkill. Too much effort for something I could just muddle through. Now the effort is negligible, so I build proper dashboards for things I used to handle with rough notes or half-remembered browser tabs:
- Comparing dishwashers, with specs and trade-offs laid out cleanly
- Travel destination shortlists with the details that actually matter when you are travelling with a baby
- Event calendars with filters so I can actually decide what to attend
These are not complicated documents. But formatted properly, they are useful in a way that notes and emails never were. The friction was the barrier. Claude removed it.
(They may not be the most useful things in the world to anyone else, but if you are curious what these actually look like, DM me. I will send you a zip folder with the files.)
Most of my research happens in the background now.
The dishwasher was just the first time I noticed this pattern clearly. Once I did, I saw it everywhere. Almost all of my information gathering, from research to comparisons to shortlists to competitive scans, now happens while I am doing something else.
I describe what I need to know, set Claude running, and get back to whatever I was working on. The browsing, the reading, the synthesising: that runs in parallel. When I check back, there is a structured output waiting.
The way I used to work: open a new tab, search, scan, lose the thread, open another tab, forget what I was looking for. An hour gone. Some half-formed notes. The way I work now: describe the question, keep working, review the output later.
My attention stays on what I am building. The data collection runs separately. That separation, between doing and gathering, is the most useful thing AI added to my day.
I have even started being more deliberate about when I use it. My Claude credits reset every Friday at 11am. Thursday night, before bed, I check my usage. Whatever is left, I put to work. I queue up four or five deep research jobs on topics I have been meaning to get to. Claude runs them overnight. Friday morning, there is a set of structured outputs waiting in the right folders, ready to read whenever I get to them. I also set up a scheduled job in Claude that runs every Thursday afternoon. It checks my usage, scans my recent chat history and work in progress, and surfaces a list of topics worth researching before the reset. I do not have to remember to do this. It prompts me.
That kind of automation, small, practical, compounding, only works because the tools I use actually let Claude in. Which is why I have become ruthless about one thing when choosing new software.
MCP integration is now my first filter for new tools.
Before I sign up for any new tool, I check one thing: does it have an MCP or CLI integration? If it does not, I move on. There are usually alternatives that do.
This became concrete when I was evaluating project management tools. I looked at several options. ClickUp had the most capable MCP integration of the ones I tested, broad enough that agents can handle the majority of tasks without me touching the UI. That was the deciding factor. Not the feature list. Not the pricing. The integration depth.
ClickUp is now wired into my Claude desktop app. I manage tasks, projects, and deadlines entirely from Claude. Every morning, a scheduled job runs across ClickUp and my calendar, compiles what needs my attention that day, and asks if anything needs re-prioritising. I read the summary. I respond if something needs to move.
If a task needs to shift to next week, I say so. Claude updates ClickUp. I never open the app.
The shift is structural: AI as the interface layer
AI as an assistant that helps you work faster is one version of this. What I am describing is different. Claude absorbed the interface layer between me and my tools. I no longer navigate applications, format documents, manage file structures, or run searches. I describe outcomes. Claude produces them.
The tools still run. I just stopped being the person who operates them. That is a structural change, not a productivity tip. And once you feel it, the old way of working does not make much sense anymore.
I work in procurement technology. I cannot close a piece like this without saying what I think this shift means for the space I spend my days in.
What this means for procurement teams
Everything I described above, the file management, the background research, the dashboards, the tool orchestration, that is a small team running a growing company. Now multiply it by a procurement team of fifty running sourcing events, managing supplier relationships, reviewing contracts, and chasing approvals across half a dozen systems.
The same structural shift is coming for them, and it is coming faster than most people expect. This is not a year-long trend. It is months. The way teams interact with their procurement tools (Coupa, SAP Ariba, Jaggaer, whatever sits in the stack) is about to change fundamentally. Agents will sit between the user and the platform. A sourcing manager will not click through four screens to build a comparison matrix. They will describe what they need, and an agent will pull supplier data, run the scoring, and surface a shortlist. A contract reviewer will not open a 60-page PDF and scroll. They will ask the agent to flag deviations from standard terms and summarise the risk. Spend analysis will not start with pulling a report and opening Excel. It will start with a question ("Where did we overspend against contract rates last quarter?") and end with a structured answer.
I have seen this firsthand. The pattern is identical to what happened with my personal tools. Once the friction of operating the software drops below a threshold, people stop operating it. They start describing outcomes instead.
Procurement teams that start working with agents and connectors now are going to pull away from everyone else. Not gradually. Rapidly.
Think about what happens when a sourcing team wires an agent into their procurement platform, their contract repository, and their spend data. Suddenly the same three people who used to spend a week pulling together a category review can have the first draft ready by morning. The agent queries the spend cube, cross-references contract terms, pulls in market benchmarks, and produces a structured brief while the team sleeps. The team reviews it, sharpens the strategy, and moves to execution. That is not a 10% improvement. That is a fundamentally different velocity.
Now think about the team down the corridor that is still clicking through the same screens, manually exporting CSVs, copy-pasting into slide decks. Same people, same tools, same hours in the day, but a fraction of the output. The gap between these two teams is going to widen every month. Not because one team is smarter, but because one team let agents handle the operating layer and the other is still doing it themselves. The teams that delay this will not just miss an efficiency gain. They will become bottlenecks. The rest of the organisation will move faster around them, and eventually the question will not be "should we adopt this" but "why haven't you."
For product teams building in this space, including us, there is an uncomfortable truth sitting underneath all of this. In the coming months, agents are going to be the largest users of our products. Not humans. That is not a prediction about the distant future. It is already starting.
This means rethinking almost everything. The UI that we have spent years polishing becomes secondary. Agents do not click buttons. The primary interface becomes the API and MCP layer. The data layer matters more than the dashboard. Integrations stop being a feature and become the product. File formats matter because agents need to read and write them reliably. I would bet that .md files become a default format within a year, simply because they are structured enough for agents to parse and light enough to pass around without overhead. Teams building procurement platforms need to start asking a different question: not "is this easy for a human to use?" but "can an agent operate this without human intervention?"
MCP support, CLI access, well-documented APIs. These are what determine whether your tools can participate in this shift or get left behind. I described this earlier with ClickUp: I picked it because its MCP integration was deep enough for agents to operate. Every procurement platform is about to face the same test from the teams that use them.
The dishwasher dashboard was a small, personal moment. But the pattern underneath it, describe the outcome, let the agent do the work, review the result, is the same pattern that is about to reshape how entire procurement organisations operate. The timeline is shorter than most people think.