9 min readBy Matt Delgado

The 2026 AI Agency Tool Stack

The stack sorted by what an agency actually does, build to support, with a named framework for telling what to run now, what to wait on, and what to skip.

Most of what gets pitched to a web design agency in 2026 is not worth a subscription. Some of it is worth a subscription today. A little of it is worth skipping outright, no matter how the demo looks. This is the stack sorted by what an agency actually does, function by function, with a rule for telling the three apart.

The framework: Run First, Wait On, Skip

Three buckets, and every tool has to earn its way into one:

Run first. It does the job today. The price is on the vendor’s own page. Nothing sits behind a beta waitlist between you and a client deliverable.
Wait on. The capability is real. One specific risk, a legal exposure, a missing review step, an unfinished workflow, makes it too early for a paying client.
Skip. The marketing has outrun the product. Or the job it claims to do belongs to a person, and always will.

Applied by function below: build, design, copy, SEO, sales, ops, and support.

Build: the coding layer

An editor that reads the whole project beats a chat window that reads one pasted function:

Tool or habit	Verdict	Why
Cursor	Run first	Cursor’s own pricing page lists Pro at $20 a month for one person and Teams at $40 a seat a month. It is an editor with a model wired into the whole repository, not a snippet tool.
A browser chatbot with no repo access	Wait on	Fine for one function pasted in and out. It cannot see the rest of the file tree, so it drifts the moment a project has more than a handful of files open at once.

Design: mockups before a line of code

A client signs off on a direction faster from a mockup than from a description of one:

Tool or habit	Verdict	Why
Figma	Run first	Figma’s own pricing page grants the free Starter plan 150 AI credits a day, up to 500 a month. The $16 a month Professional seat raises that to 3,000 a month, six times the free cap. Either is enough to sketch a real direction before anything gets built.
Skipping the mockup and going straight to a builder	Wait on	Fine for a throwaway prototype. For a client site, the companion post on AI website builders covers exactly where that tool has to hand off to hand-written work.

Copy: a first draft, not a final one

The tool writes the first pass. A person still has to make it sound like the business, not like the tool:

Tool or habit	Verdict	Why
Claude	Run first	Anthropic’s own pricing page lists the Pro plan at $20 a month billed monthly, $17 a month billed annually. Team seats start at $25 a month billed monthly, $20 billed annually. Good for a fast first pass on an FAQ page or a follow-up sequence.
Publishing the first draft unedited	Skip	It reads generic to anyone who has already seen the same generic paragraph somewhere else this week. The fix is a human edit pass, not a cleverer prompt.

SEO: structure a page around what already ranks

The tools that actually move a ranking are still the crawl, the content, and the links behind the page:

Tool or habit	Verdict	Why
Surfer SEO	Run first	Surfer’s own pricing page runs from $49 a month (Discovery) to $182 a month for Pro, its recommended plan. Every tier includes an AI writing assistant it calls Surfy. Every tier also scores content against real search results, not a guess.
Any tool promising to move rankings on its own	Skip	A chatbot wrapped around a rank tracker does not replace the crawl fixes, the content, or the links a page actually needs.

Sales: keep the lead somewhere, and call it carefully

Where a lead lands matters more than how it was drafted, and how you reach a cold prospect is a legal question, not a style one:

Tool or habit	Verdict	Why
HubSpot	Run first	HubSpot’s own product page describes a Prospecting Agent. It says the agent “monitors buying signals and launches personalized outreach automatically.” The feature ships in Starter, Professional, and Enterprise CRM tiers. A place for every lead beats a spreadsheet nobody opens twice.
AI voice cold-calling	Wait on	Federal rule 47 CFR 64.1200(a)(2) covers this. It bars an autodialed or artificial-voice telemarketing call to a cell phone with no prior express written consent. Confirm consent first. Have counsel review the script before this runs at any volume.

Ops: where the SOPs actually get read

A shared doc nobody opens is not documentation. A tool built for search and structure gets used instead:

Tool or habit	Verdict	Why
Notion	Run first	Notion’s own pricing page reserves full Notion Agent access for the $20 a member a month Business plan. That covers chat, generate, autofill, and translate. Free and Plus only get a trial. A separate Custom Agent, for a workflow you build yourself, is free to try, then $10 per 1,000 monthly credits.
Letting a Custom Agent execute a workflow with no review step	Wait on	Notion’s own page frames these as workflow-executing, not workflow-drafting. Treat that as a live risk until you have watched it run cleanly a few times on low-stakes work.

Support: a draft a person still approves

The difference between a helpful reply and a wrong one that a client sees is a single approval step:

Tool or habit	Verdict	Why
Help Scout	Run first	Help Scout’s Standard plan starts at $25 a user a month. It includes AI Drafts. The vendor describes the feature as drafting “replies based on past conversations and help content,” for a person to approve, not send.
Fully autonomous replies	Wait on	Help Scout’s own AI Answers add-on runs $0.75 a resolution. It answers with no person in the loop. A three-month free trial is the built-in way to test it, before it reaches the invoice, or a client’s customer.

Where to start, in order

Adopting all seven at once in one week is how a stack turns into subscriptions nobody actually learns:

Weeks 1 and 2: build and design. These two touch every project from the first client conversation. Get an editor and a mockup tool running before anything else on this list.
Weeks 3 and 4: copy and SEO. Add these once the build habit is set. A first-draft assistant and a content-scoring tool compound fastest once there is real client work to point them at.
Month two: sales, ops, and support. These three matter more as client count grows. A one-person shop with two clients does not need a CRM automation or a support inbox yet. A shop with six does.

Seven functions, and the pattern repeats: a paid tier with a real capability earns a run-first verdict when the vendor’s own page backs the claim, a genuine risk earns a wait, and a promise the product cannot keep earns a skip. That framework does not expire when the specific tools in this list get replaced by next year’s version of the same three buckets.