David Blank's AI Workflow Design Stack: Taste, Agents, and Tools That Actually Ship

This post started with a link cluster from David Blank, CEO at Veriti, who works across AI automation, customer experience, and workflow design. Credit to David for the weekly signal. I am adding the JQ AI SYSTEMS builder read: what these links say about how useful AI products are actually getting built.

Veriti is a drawing-native quoting product for precision CNC shops. That matters because it is a clean example of the pattern: AI becomes valuable when it captures expert judgment, keeps the human in the loop, and makes every output defensible.

Source note

I verified the public pages where possible. A few X links are treated as social signals rather than factual source material. Personal dashboard links from the notes were intentionally not published; the post links to the public Prime Intellect site instead.

1. The pattern: AI products are becoming workflow design systems

The strongest links in David's list are not random tools. They cluster into a stack:

• Industry workflow products: Veriti, BuildVision, and Ploy show AI aimed at specific business work.
• Design and website agents: Framer 3.0, Designownow, and Palmier show AI moving into creative production surfaces.
• Taste and UI systems: Kargul Studio, Awwwards, Jesse Zhou, Not Boring, Toools, 21st.dev, tweakcn, shadcn/ui, and shadcn/designer are about better taste and faster interface assembly.
• Agent app infrastructure: AI SDK, EmbedPDF, and Prime Intellect are about building real products around models, documents, evals, and routing.
• Handoff discipline: the Illustrator packaging article is a reminder that even in the AI era, missing assets and messy handoffs still break work.

The lesson is simple: AI makes the middle of the work faster, but the edges matter more. What comes in? What context is trusted? What does the user approve? What gets packaged? What can be reviewed later?

2. Industry tools: Veriti, BuildVision, and Ploy

Veriti is the best credit link to start with because it is not generic AI. It targets precision CNC quoting. The product promise is not "ask anything." It is "quote like your best estimator" by reading drawings, applying shop rules, keeping overrides, and maintaining a decision trail.

BuildVision has a similar shape for construction bid work. It reads bid packages, triages go/no-go decisions, extracts equipment, drafts quotes, and shows source references. The important design choice is that the product does not hide the judgment. It brings the decision to the human.

Ploy points at another version of this pattern: the website as an active growth system. It frames the site as something that monitors competitors, fixes technical issues, drafts landing pages, identifies visitors, and connects signals to CRM and ads. Even if you are cautious about the automation claims, the direction is correct: websites are turning into operating surfaces.

JQ AI SYSTEMS read: the best AI products do not begin with the model. They begin with a narrow workflow where the cost of missed context is high: quotes, bids, website conversion, support, compliance, packaging, reports, or client follow-up.

3. Design agents: Framer 3.0, Designownow, and Palmier

Framer 3.0 is a strong signal because Framer says agents now work on the canvas, design pages, iterate, make breakpoints, add effects, create components, write code, connect to CMS, and more. That is a huge change in posture: the website builder is no longer just a layout tool. It is becoming a collaborative site agent.

The shared Framer announcement video and the Framer community discussion both point to the same idea: AI agents are moving into the design surface with full context of styles, CMS, and site structure.

Designownow is a different but related reference: design as an on-demand operational service. It is a reminder that tools do not replace production rhythm. Clients still need briefs, requests, delivery, revisions, and taste.

Palmier brings this into video. Its useful idea is not "AI video exists." It is AI generation directly inside a timeline, with MCP support so agents can work with the project. That is exactly where creative AI needs to go: inside the edit, not in a disconnected prompt box.

4. Taste and UI systems: from inspiration to reusable components

Taste still matters. Maybe more than before.

Kargul Studio, Awwwards, Jesse Zhou, and Not Boring Software are not implementation tools. They are taste references. They remind you that a product can be useful and still have a point of view.

Toools.design is a broader directory for design resources, and 21st.dev is useful because it turns design inspiration into reusable component categories: heroes, forms, AI chats, buttons, pricing sections, tables, tabs, and more.

Then the stack gets more concrete. tweakcn helps theme shadcn interfaces. shadcn/ui describes itself as a foundation for a design system: components you can customize, extend, and build on. shadcn/designer goes further into components for building design tools: canvas, layers, drag and drop, resize, style, toolbars, and actions.

JQ AI SYSTEMS read: if you want AI-generated apps to stop looking like default dashboards, do not only prompt harder. Give the agent a design system, real components, examples, and taste constraints.

5. Agent app stack: AI SDK, EmbedPDF, and Prime Intellect

The builder layer in this link list is strong.

AI SDK is the obvious default for TypeScript builders who need streaming, tool calling, multi-provider support, fallbacks, and UI hooks. It is not just a chat wrapper anymore. It is a practical layer for building AI apps and agents.

EmbedPDF is easy to underrate. PDF interfaces are everywhere in business workflows: contracts, drawings, reports, invoices, specs, statements, policies, forms. EmbedPDF offers both a drop-in viewer and headless components, which is exactly what AI document workflows need when you want custom annotations, citations, source highlights, or review UI.

Prime Intellect points at the next layer: evaluating, training, deploying, and improving models and agents. Most small teams should not start with fine-tuning, but they should understand the direction: better agents need evals, environments, feedback loops, and deployment control.

Practical order: build the product surface first, instrument the workflow, collect examples, create evals, then consider model training. Starting with training before the workflow is clear is usually expensive theater.

6. Handoff still matters: the Illustrator packaging lesson

The most old-school link in the list may be one of the most important: ASK Design's guide to packaging Illustrator files.

It explains the boring but necessary discipline of packaging a file with links, fonts, reports, and missing-asset checks. In AI workflow terms, this is not old news. It is the same principle:

• Collect the assets.
• Preserve the links.
• Create the report.
• Check what is missing.
• Package the handoff so the next person can continue without guessing.

AI systems need the same thing. If your agent creates a design, report, quote, or app, the handoff should include sources, assumptions, files, dependencies, decisions, and open questions.

What I would try first from David's list

My personal first test would be small: build one AI-assisted document review surface with AI SDK, shadcn/ui, and EmbedPDF. Give it one workflow, one source document, one review queue, and one export. That is more useful than collecting another fifty tools.

Builder checklist

Before adding one of these tools to your stack, answer this:

1. Workflow: what repeated business job does this improve?
2. Input: what files, links, forms, docs, or emails does the agent need?
3. Context: what rules, brand styles, examples, and constraints should it know?
4. Interface: where does the human review, edit, approve, or reject?
5. Source trail: can the output point back to the file, page, rule, or user decision that produced it?
6. Handoff: what gets packaged so someone else can continue?
7. Measurement: how will you know it saved time, reduced errors, improved conversion, or increased quality?

Sources and shared links

• David Blank, CEO at Veriti, on LinkedIn
• Veriti
• Framer updates: Framer 3.0
• Framer announcement on X
• Framer community discussion
• BuildVision
• ASK Design: Packaging Illustrator files
• Ploy and Ploy on X
• Palmier
• Kargul Studio work
• Learn Framer in 20 Minutes
• Awwwards
• Jesse Zhou
• Not Boring Software
• Toools.design
• 21st.dev
• Prime Intellect
• Designownow
• tweakcn
• shadcn/ui
• shadcn/designer
• AI SDK
• EmbedPDF