What are the 5 tools you couldn’t build and work without?
These are the five tools currently doing most of the heavy lifting in my SaaS and automation workflow:
Claude — system architecture, deep product research, complex technical planning, debugging difficult problems, comparing implementation approaches, and turning unclear ideas into detailed execution plans.
Codex — working directly across large codebases, building complete features, refactoring existing systems, fixing production issues, writing tests, handling migrations, and completing changes that touch multiple files and services.
Awish — building automations and connecting the apps I use without spending hours on APIs, OAuth setup, or complicated integration steps. I use it when data, notifications, approvals, and actions need to move reliably between different tools.
Notion — keeping product requirements, technical decisions, research, feedback, and unfinished ideas organized in one place.
Slack / WhatsApp — where team communication, quick decisions, approvals, workflow updates, and important notifications actually happen.
What I’ve realized is that powerful tools alone are not enough. Most of the friction still happens between them: transferring context, checking progress, requesting approvals, and making sure the next action actually happens.
What are the five tools doing the heaviest work in your stack?
Replies
Same Claude and Codex split here, with one rule I learned the hard way: the model that wrote the plan doesn't get to grade the code. It agrees with itself every time. My unglamorous fifth is Postgres, because about half the workflows I used to wire between tools turned out to be one query and a cron, and I only worked that out after a year of paying for connectors I wasn't using.
@asadmalik901 That’s a solid rule, and I agree about Postgres. If a workflow is really just one query and a cron job, adding another platform can create more complexity than it removes.
Awish is most useful when the workflow starts crossing app boundaries — for example, reading data from one tool, updating another, asking for approval, and sending the result through Slack or WhatsApp. We’re trying to make that setup as simple as describing the workflow in plain English, without spending hours on connectors and OAuth.
I’d genuinely be curious to hear what you think of it. You can try two automations for free and see whether it replaces any of those connector-heavy workflows.
@alpertayfurr The approval step is the only part of that list I'd actually pay for. Reading and writing between apps I can do with a cron and a Postgres row, but "stop here and wait for a human to say yes" I've now rewritten three times and it's still the flakiest thing we run. Our Stripe refund flow still needs me in the loop, so that's the one I'd test it on. Where does the run state live while it's parked waiting on the approval?
@asadmalik901 That’s exactly the kind of workflow Awish is built for. While it’s waiting, the run is persisted as a paused workflow state with the completed steps, inputs, approval request, and next action saved — it isn’t relying on an open chat or in-memory agent session.
Once you approve from Slack, WhatsApp, or the dashboard, it resumes from that exact step instead of restarting the refund flow. Your Stripe case would actually be a great one to test.
@alpertayfurr Resuming from the exact step is the part most tools get wrong, they replay the whole run and you end up issuing the refund twice. The thing I'd want to break first is a stale approval: the request sits for three days, the customer has already filed a chargeback, and the saved state has no idea. If the resume just fires the next action blind, that's worse than failing loudly. Happy to run our refund flow through it this week and tell you where it snaps.