Chicago, IL · AI transformation
Chicago's premier AI transformation agency.
DRYHURST is a Chicago think tank and product studio. We take companies from AI on a roadmap slide to AI in production: strategy, governance, and the engineering to ship it, plus SaaS products built from zero.
What DRYHURST is
Strategy without build is theater. Build without strategy is waste.
Most companies face the same fork. They need a product that does not exist yet, or they need AI that works beyond a pilot. Agencies optimize for retainers. Tool vendors optimize for seats. DRYHURST optimizes for shipped systems: products in market, agents in production, decisions that hold under scrutiny.
We sit between research and delivery. On one side: clear thinking about markets, architecture, and AI readiness. On the other: engineering that leaves the lab. That dual mandate is deliberate.
Chicago · 41.88° N, 87.63° W
Chicago invented the skyscraper. We build what runs inside it.
We are based in Chicago and work with founders, CTOs and operators across the city and well beyond it. AI transformation here means one floor at a time: a use case, a production system, a team that trusts it. Click a building to switch it on.
Three ways we work
- SaaS product studioWe take an idea from concept through design, build, and launch. Founders use us when they need a serious product partner, not a body shop. Teams use us when the product must exist before the market moves on.
- AI consultingStrategy, readiness, use-case selection, governance, and adoption. We help leadership choose where AI belongs, where it does not, and what operating model keeps risk and value in balance.
- AI engineeringProduction systems: agents, LLM integrations, automation, evaluation. If it must run every day, answer for quality, and integrate with real systems, that is this lane.
Proof from products we own
We do not pad this page with unnamed logos. Our proof is work we operate ourselves, and it keeps us honest.
Invoice My Clients
Invoicing SaaS for people who need to get paid without enterprise friction, with AI and MCP integrations built in. A full product practice: ongoing shipping and the operational reality of keeping a live SaaS healthy.
OnDue
AI accounts receivable for QuickBooks Online, Xero, Odoo and Stripe Billing. OnDue reads open invoices, drafts a reminder for each overdue one and sorts what customers send back. Nothing goes out until a person approves it, and the chasing stops when the invoice is paid in the books.
Jumpstart
Takes a first-time founder from "I want to start a business" to a fully operational US company, then runs it from a single console: formation, banking, payments, books, payroll and compliance in one place.
Watch Jev decide.
Paste a support email. Jev, TypeSafe's decision model, answers five typed questions about it in one request: which team, how urgent, how frustrated, whether it asks for a refund, and whether a person should reply. No text generated, just decisions and the probabilities behind them.
This is the pattern behind our work: the model decides, code acts. Read what Jev is or how we use confidence to decide when to escalate.
Text you enter is sent to TypeSafe through OpenRouter to be evaluated, and is not stored by us. Do not paste anything private. See our privacy policy.
The lab
Don't read about it. Pull the levers.
Articles with working instruments in them: price a month of decisions, drag a probability distribution until it lies, route an inbox through two models, and translate between OpenAI's and Jev's contracts.
- 2 interactiveOpenAI's Decisions API vs Jev: Same Idea, Different ContractsOpenAI now ships a decision endpoint too. It asks the same three kinds of question as Jev with a different JSON contract and more than twice the input price. Here is the side-by-side, with an interactive translator.Contract translatorCost race
- 1 interactiveJev 1.14: Images, and a Primitive That PointsBox finder
- 2 interactiveClaude Fable 5.2 for Builders: What Changed, and What to Do About ItEffort dialMigration checklist
- 1 interactiveJev Decides, Fable 5.2 Writes: The Two-Model ArchitecturePipeline simulator
- 2 interactiveReading the Numbers: Scores, Confidence and ThresholdsScore labThreshold simulator
- Every article
Who we work with
- Founders who need a SaaS product built.
- CTOs who need AI in production, not on a roadmap slide.
- Operations leaders at SMBs and mid-market companies who want adoption with governance, not hype.
Latest articles
All articles6 min read
OpenAI's Decisions API vs Jev: Same Idea, Different Contracts
OpenAI now ships a decision endpoint too. It asks the same three kinds of question as Jev with a different JSON contract and more than twice the input price. Here is the side-by-side, with an interactive translator.
Open the article 2 interactive5 min read
Jev 1.14: Images, and a Primitive That Points
Jev can see now. Version 1.14 takes images and adds Box, a fourth question type that answers where something is. One rule makes it safe to use: a Box always returns a box, so ask whether the thing is there first.
Open the article 1 interactive5 min read
Claude Fable 5.2 for Builders: What Changed, and What to Do About It
Anthropic's most capable widely released model got cheaper and stricter. The price fell 20%, effort is now the only dial, and harnesses that edit history will get 400s. A builder's guide with an effort dial and a migration checklist you can tick off.
Open the article 2 interactive
Tell us what needs to ship.
A product, an AI program, or a production system. Send a short brief and we will reply with a direct read on fit, scope, and next steps.
Start a conversation