Data strategy and engineering - brightonSEO UK Fall 2026

Transcript

1. Data strategy & engineering for agentic workflows Ray Grieselhuber DemandSphere @raygrieselhuberl
4. Let’s talk about data strategy & engineering in an agentic world
5. Many of the same engineering principles apply in both product and data engineering
6. Today we’re building a map. Don’t worry too much about knowing it all.
7. Takeaway: A framework to help you build more reliable products
8. Our goal is to promote more fluency in building repeatable systems with AI
9. Help decision-makers understand what it takes to support AI-engineered products
10. Why am I talking about this at an SEO conference?
11. Agentic engineering can solve problems that have plagued SEO teams for decades
12. Our clients are doing things in months that would have formerly taken years
13. Examples
15. Good search marketing is good product management and vice versa
16. Performance in search is a leading indicator of the effectiveness of your corporate strategy
17. Agentic engineering enables the ability to improve both
18. Software as a Service Or Service as a Software?
19. Experience at Functionize taught me about the new SaaS
20. The dirty little secret in SaaS
21. Silicon Valley is finally starting to figure this out
22. Differentiated SaaS is an operating system, not “software”
23. This is why SaaS isn’t “dead”
24. Side note: “X is dead” is a phrase taught to startup founders to aid in fundraising
25. Vibe coding vs. AI-driven engineering
26. Don’t look down on vibe coding
27. Vibe coding is good for building prototypes and testing ideas quickly
28. AI-driven engineering is for building products
29. Vibe coding should be the starting point
30. Deployments and operations are what separate vibe coding and AI-driven engineering
31. People and processes (manual & automated) support actual products
32. Constant updates + QA (human & machine)
34. Need to anchor to business goals
35. But the cost of experimentation is cheaper
36. Let’s talk about some basics
37. Step 0: spend some time building your skills for brand, colors, assets, etc.
38. Skills like Impeccable can show you what should be stripped to avoid “Claude-isms”
39. A basic vibe coding to AI-driven engineering workflow Start in Claude.ai / ChatGPT, etc. Create repo Vanilla prototype Claude.ai / ChatGPT to build md context Move to local filesystem init Enforce testing policy Configure deployment environment Automate deployments Deploy regularly
40. The “vanilla” prototype: HTML CSS “Vanilla” JS
41. Next step: make it work
42. Have the web agent (Claude.ai, ChatGPT) build a simple working version in your target framework (Rails, Next.js, etc.)
43. Once you have a working version that the stakeholders like, have your web agent build a full product spec with context
44. Place all of these files into a project folder on the development machine
45. Create & clone your repo
46. Test both Claude Code and Codex to find your favorite
47. The apps are fine too
48. Remember the difference between vibe coding and engineering? Deployment and operations!
49. Many options for deployments
50. Go in to planning mode, plan the roadmap. M0 should be the simplest possible version (“Hello world”)
51. Place product spec and other docs in /docs Place prototype files in /prototype Set rules in context files to ensure these never make it to public viewing More importantly, have the agent build tests to prevent these sorts of leaks
52. Remember: all of these agents are very stupid liars
53. The key is to get the simplest version possible live in the production environment ASAP
54. Why? Because repeatable, automated deployments are a key factor in project success
55. Continuous Integration / Continuous Deployment
56. Just use Github Actions to build these
57. Create more rules about writing and running tests
58. Dev, staging, production
59. Exploratory testing vs. regression testing
60. Unit tests / specs, integration, browser tests
61. The faster you can get through iterations, the more successful your products will be
62. We currently deploy 5-10 times a day during sprints
63. Things we don’t have time to cover today: Database selection Host selection Cloud vs. metal Object storage Logging Security & Audits Team Collaboration Context file management APIs, MCPs, and security for these And more…
64. Final thoughts
65. Make the investment to own your own compute capabilities
66. Learn how to work with open weight models
67. Join us at FOUND Tokyo 2027
68. Thank you! (+ link to slides) https://raygrieselhuber.com