Working smarter
How To Work With AI — So It Actually Works For You
AI is the biggest skill upgrade available to anyone building or fixing websites today. But it only pays off if you work with it the right way: push back, verify, bring your knowledge — and know when to take over.
The Honest Premise
We use AI every day — in our audits, our builds, and the fixes documented in The Fix Library. And we'll tell you plainly: it is sometimes wrong, always confident, and still the most valuable tool we've added in years. Those three facts are not a contradiction. They're the job description. AI doesn't replace knowing your field — it multiplies what your knowledge can accomplish, and it punishes working without any.
Push back. That is where the correct answer lives.
When an AI's answer isn't exactly what you're seeing, say so — with specifics. "I can't find that in my HTML, where did you see it?" is not being difficult; it is the single most productive sentence in AI-assisted work. The first answer is a hypothesis built from patterns. Your pushback is the evidence that turns a plausible answer into the correct one. In our own audits, some of the best findings only surfaced after a challenge forced a re-verification against the primary source.
No single tool can troubleshoot every website.
Websites are all built differently — different platforms, themes, tag stacks, hosting layers, and years of accumulated decisions. Lighthouse sees symptoms, crawlers see structure, AI sees patterns. None of them sees your specific site whole. Treat every tool (AI included) as one witness, not the judge. When two witnesses disagree, the network panel, the actual response header, or the actual settings screen is the tiebreaker.
Know your platform — it changes the answer.
The right fix on WordPress is often impossible on HubSpot, irrelevant on Shopify, and automatic on a static site. Telling the AI what platform you're on — and what you can and cannot change there — eliminates whole categories of wrong advice before they're generated. Platform knowledge plus field knowledge are the multipliers: the more real context you bring, the better the output gets.
Verify before you act. Especially before you delete.
AI is confident by design, and confidence is not accuracy. Before acting on a finding — especially destructive actions like removing a tag, a plugin, or a redirect — verify it against something primary: DevTools, a curl of the actual header, the vendor's own documentation, the platform's settings page. A two-minute verification routinely saves a broken tracking setup or a wasted afternoon.
Embrace it where it works. Avoid it where it does not.
Used right, AI becomes an everyday tool for getting over hurdles that used to be out of scope — reading an unfamiliar error, drafting the boilerplate, explaining another platform's conventions, doing the first 80% of research in minutes. Used wrong, it confidently automates mistakes at scale. The skill is knowing which task is in front of you: delegate the leverage work, keep judgement, review, and final decisions human.
The payoff: you build skills faster, not lazier.
Working with AI the right way is a massive upgrade to your skill set and knowledge base — every solved problem teaches you the pattern, the platform quirk, and the verification method, at a pace manual trial-and-error never matched. You build better, faster, with less manual labour on the little issues technology handles easily now — and you keep compounding knowledge for the problems that still need a human.
A Real Example From Our Own Work
Auditing a HubSpot site's Lighthouse report, AI-assisted analysis flagged LinkedIn's tracking tag as installed twice — an old and a new version both downloading on every page. It looked exactly like a classic duplicate-tag mistake. The site owner pushed back: "I can't see that in my HTML — where did you see it?"
That pushback triggered the verification that should always come before a deletion: fetching LinkedIn's actual script and walking the browser's initiator chain. The verdict — LinkedIn's current tag loads its own legacy file on purpose. One installation, two files, nothing to fix. The "duplicate" was a pattern-match, not a fact. Meanwhile the same verification pass confirmed a finding that was real: two separate Facebook pixels injected at runtime by the platform, one of them stale. Push back → verify → act on what survives. That loop is the whole method, and it produced a better outcome than either the human or the AI would have reached alone. The full fix is documented here.
The Working Loop We Use Every Day
- 1
Bring context first
State your platform, what you can/cannot change, what you already tried, and paste the EXACT error or report — not a paraphrase.
- 2
Get the hypothesis
Let the AI propose causes and fixes. Treat it as a strong first draft from a fast, well-read colleague — not a verdict.
- 3
Verify against the primary source
DevTools network panel, response headers, platform settings, vendor docs. Confirm the finding exists on YOUR site, not just in the pattern.
- 4
Push back with what you found
"That file isn't in my HTML" / "my platform has no such setting" — every specific correction narrows the search and upgrades the next answer.
- 5
Act on the verified subset
Apply what survived verification. Start with reversible changes; save deletions and structural changes for last.
- 6
Retest and keep the lesson
Re-run the audit or test, confirm the outcome, and note what was true for your stack — that knowledge compounds into the next problem.
Where To Draw The Line
Lean on AI for
- • Decoding unfamiliar errors and reports fast
- • First drafts: code, copy, configs, checklists
- • Explaining another platform's conventions
- • The first 80% of research and triage
- • Tasks outside your usual scope — that's the upgrade
Keep human
- • Verification against primary sources
- • Destructive actions: deletions, DNS, redirects
- • Anything below your quality bar — redo it manually
- • Judgement calls that need business context
- • The final review, every time
Together, that combination — AI leverage plus human review and troubleshooting — consistently produces a better outcome than either alone. Embrace it where it performs, step in where it doesn't meet your standards, and it becomes a tool you reach for every single day.
Reach the Summit
Want a team that already works this way?
We combine AI-assisted speed with human verification on every audit, build, and fix — that's how The Fix Library gets written.
Start the Conversation