Back to case studies
LlamaIndex · 2024–2025
AIDev toolsB2B startup0→1 design team

Building the design function at LlamaIndex

As the first designer at a pre-launch AI dev-tools startup, I built the design practice — moving it from a late execution step toward a partner in product planning.

Role
Founding Product Designer
Team
CEO · CTO · Head of Product · Head of DevRel · 8 Engineers
Outcome
Operating model in use · Design function established
Design-maturity assessment
my baseline on joining
Not in place
0123
Established
Operational33%
  • Design leader0
  • Trust credit1
  • Quick wins1
  • T-shaped specialists2
  • Hiring0
  • Task planning1
  • Shared tooling1
  • Design critique2
  • Trusted outsourcers1
Tactical9%
  • Org structure0
  • Decision-maker map1
  • Teamwork skills1
  • Growing designers0
  • Marketing design wins0
  • Design system0
  • Design debt0
  • Checklists0
  • Standard methods0
  • Meeting user needs1
  • Prototypes0
Strategic6%
  • Company design literacy1
  • Co-design1
  • Design-leaders club0
  • Long-term planning0
  • Influence on roadmap0
  • Business-tied metrics0
  • Discovering user needs0
  • Customer Journey Map0
  • Knowledge & insights base0
  • ResearchOps0
  • Brand–interface link0
  • Employer brand0
One-year goal Lock in the operational floor, then move the practice up to tactical — design integrated into product planning, not bolted on at the end.

I joined the company before launch. The MVP had been built by the engineering team without a designer involved, and there was no research or QA process in place. Before starting on screens, I ran a diagnosis — where the design practice stood, who influenced the product, and where design could enter the process earlier.

What changed
Design-review gate
before every release, agreed with engineering
15%
of each sprint committed by engineering to design debt
Notion Linear
whole team migrated onto one planning system I set up
PURE baseline
usability scored across the 7 core product flows
01  ·  The problem

Design had no process, no research, and little trust as a function.

Tasks arrived at the last minute, engineers changed the design on the fly, and design steps were skipped whenever speed was the priority. Design was brought in to execute at the end — rarely to help decide what got built.

01
No process or planning

No formal way to brief design, no mid-term planning, and design routinely dropped from the schedule to hit a deadline.

02
No research or QA

No researchers, no QA process, and no reliable way to recruit respondents — so design decisions couldn't be validated before they shipped.

03
An inconsistent product

shadcn was applied unevenly — the same components looked different across the product, so the product looked unfinished.

04
Low trust credit

Design wasn't seen as a driver of quality or value, and decisions defaulted to engineering over UX.

The product works with sensitive data and has to deliver accurate results, so an inconsistent, unreliable experience directly affected how much users trusted it and wanted to use it.
02  ·  The approach

Diagnose before designing: measure the practice, map the people, then plan how to change the process.

I treated the design function like a product with a maturity problem. Before changing how the team worked, I scored where the practice stood, audited the resources, mapped who influenced the product, and made an explicit plan for building trust.

  1. 01
    Score the maturity. Rate the practice across three tiers — operational, tactical, strategic — to set an honest baseline and a one-year target.
  2. 02
    Audit the resources. Money, team, and time — separately, because they were not equally scarce.
  3. 03
    Map the decision-makers. Plot every key person by influence and interest, then set a communication cadence per quadrant.
  4. 04
    Read the business priorities. What the company needed from design now vs. in a year — so design bets matched where the business was heading.
  5. 05
    Plan the trust credit. Decide deliberately how to build credibility, and how not to lose it.
Diagnosis · 01

Auditing money, team, and time separately

The company had money; what it lacked was people and time. So the plan couldn't rely on hiring alone — it had to use the existing time better.

MoneyEnough
  • Tools and licences
  • Room to invest in designers
  • Employer brand
TeamShort
  • One product designer
  • No researchers
  • Too few front-end engineers to implement well
TimeShort
  • No time for detailed flows
  • No time for research
  • No time to apply review notes
Diagnosis · 03

Mapping the decision-makers by influence and interest.

Influence and interest didn't line up: the people with the most influence over design weren't always the most interested in it. I plotted each decision-maker on a 2×2 and set an engagement tactic and cadence for each quadrant. Two groups mattered most — the key partners (PM and CTO) to involve from the start, and the high influence but low interest (engineers), who could hold up delivery if not kept in sync.

High interest Low interest
Low influence · interested
Head of Product

Wants a strong result and values an outside view — but can read design input as noise, so weighs proposals critically.

Keep in the loop · Weekly design sync
High influence · interestedKey
Product Manager · CTO

Directly shape the design. The key managers — involve, align, and co-decide with them fully.

Involve from day one · Daily syncs
Low influence · not interested
CEO

Low day-to-day involvement. Keep informed; surface design's value through shipped results.

Keep informed · Town Hall after first alpha/beta
High influence · not interestedWatch
Engineers

Can hold up on-time delivery if out of sync. Sync regularly and offer a comfortable level of involvement — e.g. aligning on guidelines.

Async in Slack, weekly
Low influence High influence
Same 2×2, two overlays: where each person sits (influence × interest) and how I'd work with them (tactic + cadence). Merging the map and the communication plan kept the strategy on a single page.
Diagnosis · 04

What the business needed from design — now vs. in a year

Mapping the business's priorities over time showed where to place design bets. The two rows that flip → are the shift I planned around: the company was moving from polish now toward speed and building a design team.

Value to the businessNowIn a year
Product qualityCriticalTolerable
Speed to market flip →TolerableCritical
Lowering launch riskTolerableTolerable
Growing audience & revenueCriticalTolerable
Strengthening the brandCriticalTolerable
Building the design team flip →TolerableCritical
Recognising design's valueTolerableTolerable
Diagnosis · 05

Trust credit: what earns it, and what burns it.

With low credibility at the start, every move either added to or drew down design's credit — so I made both sides explicit to guide the team's early choices.

Earns credit
  • Quick, visible wins that solve real customer problems
  • Design in the room from planning, not just execution
  • Research insights presented — wins tied back to interviews
  • Decisions explained to non-designers; results shown at Town Hall
  • Designers owning scope, timelines, and outcomes
Burns credit
  • Missed deadlines and silent slips
  • Opaque work — no one knows what design is doing or why
  • Sloppy detail: broken standards, errors in final mockups
  • Shallow understanding of the product
  • Design left to last, with no influence on strategy
07  ·  Operating changes

Four operating changes, starting with quick wins.

High value and low effort first — to build trust — before moving to the deeper changes.

1

An audit, ranked with the product team

I audited the whole app and logged every problem, then ran a 2×2 prioritisation with the product team and turned the top items into sprint tickets — so the backlog was shared, not imposed.

2

A design-review gate + 15% for design debt

Engineering agreed to a design review before every release, and to spend 15% of each sprint on design-debt tickets. That agreement gave design a regular place in the release process.

3

Support tickets as a research proxy

With no research budget, I monitored post-launch support requests to surface the most frequent problems — a cheap signal that pointed the roadmap at what users actually hit.

4

A workshop series to co-design with decision-makers

To build trust with the team, I ran short decision-maker workshops: brief, Q&A, then everyone drafts a concept solo in Miro (group discussion tends to create noise). I synthesised the concepts into one report the team debriefed — widening the range of solutions while involving stakeholders in the design.

Instead of design working on its own — the people with influence but low interest became participants rather than blockers.

Operating model

Migrating the team from Notion to Linear.

Sprint work lived in Notion — issues, planning, everything. It had grown disorganised: no shared way to write tasks, stale state no one kept current, and the tool lagged under the volume of content. I proposed and ran the move to Linear end to end.

Before · NotionDisorganised
  • No shared standard for writing tasks
  • Stale state — impossible to tell done from open
  • Performance lag under the volume of content
After · LinearA system
  • Team, projects, sprints, tags, views, task template — all set up
  • Everything migrated across from Notion
  • Slack integration + team onboarding workshop
What it changed
Task state stayed current

The team kept issues up to date — so current load became visible and sprints got easier to plan.

The backlog became workable

A deliberate tag system (Customer · Scope · Bug) made the backlog easier to navigate.

Support fed the backlog

The Slack integration made it easy to turn incoming support into tasks, and made task changes visible to everyone.

I also wrote the Linear task template the team briefs against — goals & success metric, the design tasks, research/analytics that motivated it, the proposed solution, constraints, and labels — so every request arrived with the context design needs to start.
Measuring quality

A usability baseline across the seven core flows.

To make quality something we could track rather than argue about, I scored each core flow with the PURE method (Pragmatic Usability Rating by Experts). Higher means more friction — which turned the two heaviest flows, Create new Index and Parse file via API, into the first fix targets.

Easy Moderate High friction
FlowFrictionScore
Initial enrolment
3
Create new project
1
Create new Index
6
Sync Index
3
Create API key
2
Parse file in the UI
3
Parse file via API
5
Scores are my own PURE ratings per flow. The baseline gave the roadmap an objective starting point and something to measure re-tests against.
10  ·  Outcome

Design moved from the end of the process to a regular part of it.

Back to the baseline I opened with. The one-year goal was to lock in the operational floor and push the practice into tactical — design in product planning, not bolted on at the end. Re-scoring the same three tiers a year on:

Operational
the floor
33 78%
Tactical
the goal
9 45%
Strategic
next horizon
6 18%
Baseline on joining One year on
The operational floor was secured and the practice crossed into tactical — the goal — while strategic work (roadmap influence, business-tied metrics) is the horizon for year two. Scores are my own self-assessment against the same three-tier model.
01

Design had a regular place in the release cycle — a review before every ship, plus a 15% sprint allocation to work down design debt.

02

The team ran on one planning system with current state, a workable backlog, and support flowing into tickets via Slack.

03

Decisions started earlier: a shared audit, a stakeholder plan, and co-design workshops made design a partner in planning, not a service at the end.

04

Quality became measurable — a PURE usability baseline across the seven core flows, and a maturity target to move the practice toward tactical.

What I'd carry forward

Three habits I set out to build into the team

Design is the whole team's job

Everyone — designer, engineer, QA, PM, marketer — watches the quality of the experience and can propose an improvement. Design stops being one person's bottleneck.

Keep the work in the open

Visible process builds trust. Show intermediate work, tie wins to insights, and make it clear what design is doing and why.

Stay close to the user's problems

Notice what users struggle with in every interaction, write it down, and bring it back to the team — even when there's no formal research budget.

The stronger an organization's culture for design, the greater its commitment to using design as a resource. Businesses need to develop their own design culture by applying design leadership and applying effective design-management practice … Only then will the idea of design as a business resource achieve its true power as a means of reaching business objectives.
Thomas Lockwood · Design Management Review
A design-management case. The maturity percentages are my own internal self-assessment scores against a three-tier model; other outcomes are described qualitatively. No confidential company data is shown.
Next case
Making data lineage legible