About a year ago I wrote up the SaaS-to-custom-tools swap — the original $2K/month bill, what I replaced, what I kept, what the API line item looked like once the dust settled. The honest answer at the time was the monthly run-rate dropped to around $75-105 and I'd spent four or five months of part-time work to get there.
This is the same writeup, twelve months later. What got killed in the year since. What I quietly added back. What I built that didn't exist when I wrote the first piece. And how the monthly number actually moved.
What I Replaced in 2026
The original article covered the obvious cuts — reporting dashboards, creative tools, landing pages, SEO platforms, audience research. Most of those replacements are still in place and have only gotten better. The new cuts in the last twelve months were the second-tier subscriptions, the ones I'd justified keeping because the custom alternative wasn't sharp enough yet. By early 2026 the alternative was sharp enough.
Looker Studio + the freelance report builder
Replaced: Looker Studio dashboards + ~$200-300/month freelance weekly report builder
Built: monday-report + weekly-rollup Claude Code skills
Current cost: ~$10/month in API + Resend send fees
Looker was the last reporting dashboard I hadn't killed. I kept it because it's free and clients sometimes asked for "a dashboard." The problem is nobody actually opened the dashboards. The Monday email is what they opened.
So I built the email natively. The monday-report skill pulls Meta, Google, Shopify, and Amazon for the past week, renders an HTML email with the daily breakdown table and product mix, and sends it via Resend with the client on copy. It looks like a thoughtful note because the framework is one — the skill fills in the numbers. I wrote about the broader reporting stack architecture in a separate piece.
The freelance report builder went too. That was a $200-300/account/month line item where a contractor took my numbers and made them pretty. The skill makes them pretty.
Creative SaaS leftovers (the Canva tab, the prompt galleries)
Replaced: Canva Pro + a couple of AI prompt-template subscriptions (~$30-80/month)
Built: imggen CLI workflow + the Creative Engine skills (hook-battery, ad-copy, creative-brief)
Current cost: API fees only — folded into the existing $15-30/month
The original article killed AdCreative.ai and most of the design SaaS. What I quietly kept was Canva, mostly for one-off resizes and the occasional "I need a quick social tile" task. By the end of 2025 the imggen CLI plus a brand-config file handled every one of those use cases without a browser tab open.
The bigger replacement in 2026 was the prompt-template subscriptions I'd been paying for. Two of them, both in the $15-40/month range, both selling "winning ad prompt libraries." The hook-battery and ad-copy skills produce better output because they're grounded in RMBC frameworks and the specific brand context, not a generic library. Same with creative-brief — it eats a product brief and outputs a structured brief I hand to the design pipeline.
Copy and outline tools
Replaced: A long-form copy assistant subscription ($50/month) + a separate landing-page wireframe tool ($30/month)
Built: lander-copy, copy-rewrite, email-promo, and the broader RMBC skill family
Current cost: Claude API only — already counted in the monthly total
I'd been paying for a copy assistant that was essentially a templated prompt wrapper around an older model. Once Claude 4 landed and the RMBC skills matured into a 40-plus skill library, the wrapper had nothing left to offer. lander-copy writes the page. copy-rewrite audits and revises existing copy with a before/after diff. email-promo writes the broadcast. The wireframe tool was the same story — most of what it did was reorder sections I'd already structured in the brief.
Audience research subscriptions
Replaced: A persona-research SaaS I'd been testing (~$80/month) + holdover SparkToro renewals on two accounts
Built: unified-research-synthesizer + the Intelligence Suite-style market research pipeline
Current cost: ~$3-5 in Claude API per full research pass
The original article replaced SparkToro for my main book. What was left was a couple of legacy client accounts where the renewal kept auto-charging, plus a newer persona-research tool I'd been testing for B2B work. Both got cut once unified-research-synthesizer could combine reviews, Reddit threads, competitor positioning, and customer voice into a single document I'd actually use.
This one wasn't a huge dollar cut — maybe $80-100/month total — but it removed three more logins from the rotation.
What I Kept
Not everything had to go custom. The point was never to zero out the SaaS bill, it was to stop paying $200/month for a UI on top of a free API. Some tools earn their fee. These are the ones still in the stack:
- Stripe — payments. There is no custom alternative. ~2.9% + 30¢ per transaction, no monthly fee.
- Resend — transactional email send infrastructure. ~$20/month at the volume I run. Could swap to AWS SES for less but the API ergonomics aren't worth the savings.
- Supabase — split-test admin + auth for a couple of internal dashboards. Free tier handles most of it; ~$25/month on one project that outgrew it.
- GitHub — code hosting + GitHub Pages for landing pages. Free for the workflow I run.
- Vercel — hosts the Operator's Stack site + a couple of client funnels. Free tier covers the small stuff.
- 1Password — credential storage. ~$8/month. Boring and essential.
- Cloudflare — DNS and the occasional WAF rule. Free tier.
- Anthropic API — the actual model calls. This is the bill that grew.
The pattern: infrastructure stays, dashboards go. Anything that's purely a UI on top of an API I could call myself got replaced. Anything that's actually hard infrastructure — payments, deliverability, auth, DNS — stays bought.
The Build Cost
The original article said the first wave of tools represented roughly 4-5 months of part-time development. The second wave in the year since was lighter — call it another 200-250 hours spread across twelve months, mostly evenings and one or two focused weekends. Some of that was new skills, some was rewriting earlier tools as Claude got better and the patterns matured.
Where it paid off: the reporting skills. Setup time per client dropped from "an afternoon" to "thirty minutes." The Monday report skill alone saves a couple of hours per week per account. The creative pipeline pays off every time a campaign needs net-new variants.
Where it didn't pay off as cleanly: a couple of one-off skills I built for specific workflows that I ended up using twice and abandoning. The hidden tax of a custom stack is the skills that don't survive contact with real work. Maybe one in five of the things I built in 2025 is still in active rotation. The other four either got rolled into a better skill or quietly deleted.
The Monthly Delta
Here's the updated table. Original article numbers on the left, twelve months later on the right.
| Category | March 2025 | May 2026 |
|---|---|---|
| Reporting / Attribution | $400 → $0 | Still $0 |
| Ad Creative Tools | $300 → $15-30 (API) | $15-30 (API) |
| Landing Page Builder | $150 → $5 (Supabase) | $5-25 (Supabase) |
| SEO Platform | $300 → $50 (DataForSEO) | $50 (DataForSEO) |
| Audience Research | $150 → $0 | $0 |
| Looker / freelance reporting builder | (not in original) $200-300 | $0 |
| Copy / persona / template subs | (not in original) $80-150 | $0 |
| Canva + one-off creative SaaS | (not in original) $30-80 | $0 |
| Resend (transactional email) | (negligible) | $20 |
| Anthropic API (model calls) | $10-20 | $80-150 |
| Monthly total | ~$75-105 | ~$170-280 |
The headline is the total went up, not down. That's not failure — it's the API line growing because the skills got bigger and the volume went up. Twelve months ago Claude was a copy generator. In 2026 it's running the reporting, the research synthesis, the creative briefing, the keyword classification, and most of the campaign-build work. The Anthropic bill replaced a stack of subscriptions that would have cost three or four times more at this usage level.
Pulled the other way: the SaaS line items I cut in 2026 — Looker + freelance reporting + the leftover copy and research subs — would have totaled another $400-600/month if I'd kept them. So the net versus "if I'd done nothing" is still strongly negative. The custom stack is still cheaper by a wide margin. It just isn't $100/month cheap anymore.
What's Harder Than I Expected
Three things I underestimated when I wrote the original piece.
Maintenance load is real. Meta versioned the Graph API twice this year. Each time, three or four skills broke quietly — they'd run, return zeros, and I wouldn't notice until a Monday report looked off. The fix is usually small, but the noticing isn't. I now have a smoke-test skill that pings each API once a day and yells when the response shape changes. That's another thing to maintain.
API pricing drifts. DataForSEO raised on one endpoint, Resend changed their free tier limits, and one of the image model providers I was using doubled its per-call cost. Custom stacks don't insulate you from this — they expose you to it more directly than SaaS does, because SaaS dashboards eat the change on your behalf.
Skills rot. The skills I wrote in early 2025 against an older Claude model needed rewrites once Claude 4 landed. Not because they broke — they just got dramatically better when rewritten with more reasoning headroom. That's a tax I didn't anticipate. Useful tax, but a tax.
What's Easier Than I Expected
Three things that surprised me the other direction.
Reasoning gains in Claude 4 made entire categories of skills trivial. The unified research synthesizer would have been a multi-week build in 2024. In 2026 it was a long weekend. The classifier that decides which articles to write next, the auditor that scores RMBC compliance, the synth that merges five research docs into one — all easier than they would have been twelve months ago.
The skill ecosystem matured. When I wrote the first article, I was building skills from scratch every time. By 2026 there's a body of patterns — slash command shape, shared context modules, common helper scripts — that means a new skill is mostly composition. The marginal cost of skill number forty is much lower than the marginal cost of skill number five.
Client onboarding got fast. A new account used to take a day of plumbing — API access, pixel verification, brand config, reporting wiring. In 2026 there's a client-onboarding skill that scaffolds the whole folder, pulls baseline research from the URLs and handles, and stages the config files. The first hour does what used to take eight.
The Bigger Picture
Twelve months in, I don't think the math changed. SaaS bills for marketing tools are still mostly dashboards on top of APIs, and most of them are still replaceable by a few hundred lines of orchestration. What changed is the ratio. In 2025 it felt like a heroic effort to replace a $200/month dashboard with a skill. In 2026 it's a Tuesday afternoon.
The new constraint isn't whether you can replace the SaaS. It's whether you should — whether your workflow is specific enough to benefit from custom, and whether the maintenance tax fits your operating model. For a solo operator or a small team with technical comfort, the answer is still yes. For a non-technical marketing team, the answer is probably still no.
If you want the actual patterns instead of guessing (the API wiring, the skill shape, the reporting and creative workflows), The Operator covers them end to end: pre-built Claude Code skills you install and run, plus The Lab where new skills land every month. One-time $397 launch price, going up as the Lab grows. Or hire Clare Digital and we'll run this stack across your accounts.