Claudio Lassala's Blog

Last week, I attended Jake’s excellent talk at the Agile Leadership Network: AI Enhanced Productivity for Agile Teams. Here are some thoughts that came to mind (you’ll find a link to the recording at the end of this post).

What Sparked This Post

Jake and I sit four desks apart, yet we’ve never worked on the same project. Hearing his talk made me smile because it overlaps a lot with a talk I’ve been giving: Will AI Take Our Job, or Join Your Scrum? We converge on the same conclusion: AI is an accelerant for teams that practice the fundamentals well.

---

Communication → Code

I’ve seen a consistent pattern for years: developers who struggle to express ideas verbally or in writing frequently produce confusing code. It’s not a character flaw; it’s a feedback loop. Vague thinking leads to vague sentences, which become vague names, vague models, and vague tests.

AI doesn’t fix that. It amplifies whatever clarity (or murkiness) we bring to it. If my prompt is fuzzy, the output may be polished, but it’s still fuzzy in the ways that matter.

---

Three Ways to Think Out Loud (and Feed AI)

When words stall, pick another lane. Modern tooling lets us move fluidly among writing, speech, and visuals:

1. Write it. Draft the intent, constraints, and examples. Tighten verbs and nouns. Replace adjectives with concrete acceptance criteria.
2. Say it. Dictate your thoughts; let AI transcribe. Then edit the transcript into a spec. Speaking first is often faster than staring at a blank page.
3. Draw it. If you can’t find the words, sketch the flow: boxes, arrows, states, and boundary conditions. Or, sketch a feature’s mockup. Then describe the drawing to AI (or attach it when possible) and ask for a first-pass spec.

I regularly hop across all three to get from a fuzzy idea to a usable specification.

---

Spec‑Driven Development Isn’t New

People are excited about “spec‑driven development.” Great! But many of us have been here for a while. When I moved from TDD to BDD, we stopped calling them tests and started calling them specs. Specification by Example and Given‑When‑Then gave us a durable habit: describe behavior first, then implement.

That’s why AI has been smoother to adopt for teams already practicing BDD; the specs are already there. AI has something solid to work from.

Example (concise GWT):

Feature: Pricing rules
  Scenario: Apply customer discount to eligible items
    Given a customer with a 10% discount
      And an order with two eligible items totaling $100
    When the order is priced
    Then the final price is $90
      And the discount is recorded on the invoice

That little block is gold. It clarifies intent, exposes edge cases, and gives AI exactly the context it needs to scaffold code, tests, and docs without drifting toward mediocrity.

---

Practical Routines You Can Adopt Today

• Start from a spec: Before opening an editor, capture a single GWT scenario (or two). Ask AI to propose edge cases you missed.
• Dictate → Spec: Talk through the feature for two minutes. Convert the transcript into acceptance criteria.
• Sketch the flow and/or draw a mockup: Sequence diagrams, state charts, or a quick box‑and‑arrow. Feed the image (or a description) to AI; request a draft API contract and failure modes.
• Name things with intent: Have AI propose 10 alternate names for a murky type/method. Choose the clearest; rename everywhere.
• Log decisions: Keep lightweight ADRs (Architecture Decision Record). Paste key ADRs and domain glossary into the repo so coding agents stay on the rails.
• QA moves left: Invite QA to attack the first AI‑drafted spec with edge cases before a line of production code is written.

---

Where Jake’s Talk Overlaps My Own

• AI as teammate, not replacement. Humans hold the why and the trade‑offs.
• Smaller units, faster loops. Clear, measurable slices enable AI to assist without overwhelming us with tech debt.
• Non‑negotiables: tests, living docs, CI/CD, and human review remain the backbone.
• Security & governance: Pay for enterprise‑safe tools; sanitize inputs; set guardrails; vet vendors.

Different paths, same destination: disciplined teams get more creative time because AI clears the boilerplate.

---

Closing Thought

If communication shapes code, then improving how we write, speak, see, and draw is one of the highest‑leverage ways to improve our software, especially with AI in the loop. Start with one feature this week: write a tiny spec, talk it out, or sketch the flow, and let AI meet you there.

---

Watch Jake’s ALN’s AI Enhanced Productivity for Agile Teams talk here.