How AI Agents Are Changing the Data Consultancy Game w/ Chris Tabb (Confluent Current London 2026)

TL;DR

• 1

  Agentic workflows are reviving old engineering discipline — Chris Tabb says the jump from “vibe coding” to useful multi-agent systems came from bringing back requirements, definition of done, review loops, and specialist handoffs across the SDLC.

• 2

  Local and open-weight models are becoming the practical choice — Frustration with premium-model limits like Claude usage caps is pushing teams toward on-prem or local models that are cheaper, tunable, and often “good enough” with strong prompts and guardrails.

• 3

  Consultancies now compete on reusable skills, not just billable people — At Elite Data, Tabb is building internal utilities, standards, and “tribal knowledge” so agents can consistently generate data models, dbt patterns, DDL, and security work across clients.

• 4

  Bad incentives create pointless token burn — Tabb calls out token leaderboard culture at big tech firms, where people gamed usage metrics by running unnecessary jobs, like using premium models for trivial tasks such as spell-checking.

• 5

  Prompt management is really context management — One of Tabb’s biggest lessons from building content and coding utilities is that tuning prompts on bad context is a dead end; the input hierarchy and retrieved context matter just as much as the wording.

• 6

  AI removes excuses for skipping data modeling, but not the need to know what “good” looks like — Joe Reis and Tabb argue that agents can eliminate data-modeling bottlenecks, yet humans still need architectural judgment to prevent monoliths, inefficiency, and badly structured systems.