My Freelance Stack Was a Distributed System and Nobody Was Paging Anyone

I am a backend developer. I think in queues, idempotency keys, and eventual consistency. So when I tell you that the operations layer of my freelance business looks like a distributed system that has been running without observability for three years, I am not being dramatic. I am being precise.

Here is the actual stack as of the start of this year. Time tracking in Toggl. Invoicing in a small EU-based tool I will not name. Bookkeeping in a spreadsheet that exports to a CSV that I email to my accountant. Project management in Linear for some clients, Asana for others, Notion for a third group. Payments arriving across two bank accounts and one multi-currency wallet. Receipts in a folder labelled "receipts" that I have not opened since November. Tax provisioning in a separate spreadsheet that I update from memory.

The state of the system is recoverable but not reliable. The signal is in there somewhere. The pipelines do not reconcile.

The cost of this is not the apps. It is the cognitive overhead of moving data between them. Every Friday I copy hours from Toggl into invoices. Every month I copy invoice totals into the bookkeeping sheet. Every quarter I reconstruct cash flow from three sources because no single source has the full picture. Every year I spend an unholy weekend in March pretending I will reform.

A comparison of Airtable, Notion and ClickUp as all-in-one tools for freelancers finally articulated something I had been circling without naming. The all-in-one promise has historically failed because the tools that try to be all-in-one are mediocre at any one thing. The ones that are best in class at one thing are deliberately not all-in-one because that would dilute their depth. So we end up with a stack of seven excellent tools that do not talk to each other, or one mediocre tool that pretends to do everything and frustrates us into a stack of seven excellent tools.

The article does not pretend to solve this. It frames it. That alone was useful. I had been blaming myself for not finding the perfect tool. There is no perfect tool. There is a set of trade-offs and a quarterly review.

What I have done, after sitting with this for a while, is the following. None of it is novel. All of it works.

I picked one source of truth for hours and one source of truth for money. Hours live in Toggl. Money lives in a single bookkeeping ledger. Everything else is downstream of those two. If a tool cannot import from those two or export back to them, it does not enter my stack.

I built a Sunday evening reconciliation routine. Forty-five minutes. I review the week's hours, generate any invoices due, log any payments received, and capture any uncategorised expenses. The routine is non-negotiable. I miss it perhaps two weekends a year. The cost of missing it is roughly two hours of catch-up the following week, which is a strong incentive on its own.

I stopped trying to automate the messy parts. The parts that look automateable, like categorising expenses, are exactly the parts where my judgement is the thing being automated, and removing the judgement creates errors that are expensive to find later. So I do them by hand on Sunday, which sounds backwards but is actually faster end-to-end than fixing the mess every quarter.

There is also an honest tour of the best all-in-one productivity apps for freelancers that lists the trade-offs across the main contenders. The honesty was the part I valued. Most reviews of these tools are sponsored or shaped by affiliate incentives. This one was usable as decision input rather than entertainment.

What I want, eventually, is a system where the act of completing work emits the right events, and the right downstream artefacts get generated automatically. The contract closes a Linear issue, which writes hours to the ledger, which generates an invoice on the cycle date, which lands in the wallet, which reconciles against the bookkeeping, which provisions tax, which projects cash flow. That is a real distributed system, with proper boundaries, owners, and observability. We do not have it yet. We will, probably. The architecture exists in pieces across many vendors.

For now, the Sunday routine and the two sources of truth are the boring discipline that makes the rest tolerable. I am not data-blind anymore. I am data-late, which is closer to manageable.