Research Tech Retrospective
Post-engagement retrospective from the first full Zebra sprint.
Research Tech Sprint Retrospective
3 sprints / 25 days / January 2026
Decisions at the end. Everything else is evidence.
What Went Well
The Product Works
Client reaction says it all: "I'm amazed. Unbelievable." "I can't wait to give it to users." Immediate ship plans, not polite appreciation. Permission to quote on website — locked in on the call.
55k lines of code. 173 components. 25+ screens. Visual design scored 9/10 — Linear, Notion, Raycast tier. €15k paid for €40-65k standard market value.
This validates that frontend design engineering works as a productised sprint process. The natural evolution of the Deep Work Hypersprint — updated for an era where you ship code, not Figma. The methodology lineage is real: GV (1.0) → AJ&Smart (2.0) → Deep Work (3.0) → Zebra (4.0). Each version learned from the previous. This one ships production code.
The React stack worked well for me. Reuse it: React 19 + TanStack Router + Tailwind CSS v4 + shadcn/ui + Vite.
User Testing Is the Differentiator
9 interviews across 2 rounds. ICP crystallised from "undefined and outdated" to two clear segments. Session duration halved between rounds (60 min → 25-30 min). Feedback shifted from "what is this?" to "I want more depth."
The "chat reduces trust" finding alone saved the client months of engineering in the wrong direction. 7 strategic decisions validated with real user data, not opinion.
This is what makes Zebra exceptional as design and code become commoditised. Anyone can build a website now. Not everyone can test it with real users, weight the results by ICP fit, and iterate same-week based on evidence. User testing in person, structured and rigorous — this is the thing that sustains premium positioning when AI makes implementation table stakes.
Workshops Create Boundaries That Work
Zero feedback rounds on this project. The clients accepted the sprint results as the results. Best example: final presentation — Bartek asked if he could add scope, knowing he couldn't, and accepted "no" without a single pushback.
The workshop → build → validate cycle creates structural boundaries that protect both parties. The process says no, not you. This works for clients (clarity, professionalism) and for AuDHD needs (no ambiguous scope conversations mid-sprint).
Kamil: "I like the workshop approach — it's a very good approach to get the client requirements."
Prototype-First Feedback Is the Zebra Approach
Sprint 2 workshop on working code was far more productive than sketching would have been. Clients reacted to real interactions, decisions were concrete. From Deep Work experience: designer work got upvoted anyway, client sketches added "small details, if any."
In an AI coding world, getting to working software and iterating from there beats paper sketches. Code before consensus.
Day 3 Scope Lock Is a Win, Not a Workaround
Workshop 1 didn't lock scope — and that's fine. The team debated ICP, flagged outdated research, explored direction. That's valuable. The insight: Workshop 1 creates alignment on direction. Day 3 creates scope lock.
The process: Workshop 1 (align) → 2 days of research and building → Day 3 sync (present scope, visual design, lock in with the team). This worked exceptionally well and is a better process than Deep Work's approach. Standardise it as methodology, not as a fix for a failure.
Supervote Works for Feature Prioritisation
Quick, efficient, no endless debate. Decider has final say but hears everyone's reasoning first. Standard for any "what do we focus on" question.
Category of One — Confirmed
Comprehensive market research: no competitor offers the full combination of senior UX expertise + sprint methodology + ships to client codebase + user testing as endpoint.
| Category | What They Don't Do |
|---|---|
| Design Sprint agencies (AJ&Smart) | Ship code |
| Subscription design (DesignJoy) | Code, testing, strategy |
| Lovable agencies | Ship to client codebases |
| AI MVP shops | Senior UX expertise, validation |
| Fractional design | Implementation, code, testing |
12-18 month window before the market catches up. This aligns with the Oversubscribed principle (Daniel Priestley) — creating a market of one. The deeper we go into this specific combination, the stronger the positioning becomes. Seed-stage startups on a funding timeline don't have alternatives. Hiring takes 5-9 months. Agencies deliver Figma in 12-16 weeks. The Zebra Sprint is the only viable path that fits the timeline.
Referral Loop Started Organically
Bartek offered 25% referral fee unprompted. Warm lead conversion worked exactly as predicted — Szymon introduction, easy yes to €15k, 50%+ close rate pattern confirmed.
What Needs Solving
Energy Collapse Is the Priority Problem
Final sprint ended at 2am. Exercise dropped entirely in the last week. Days skipped, then compensated with unsustainable pushes. Every core boundary from the business plan was violated: max 4-5 hours deep work, energy threshold below 7/10, exercise as non-negotiable.
Charlie on the call: "I wouldn't be able to maintain this week by week. I'm going to have to take next week off."
This is the most important thing to solve. But it needs honest context:
The business plan says "if any revenue target threatens the working pattern, the pattern wins." That's the long-term principle. The short-term reality is that until family financial security is established (€30k reserves), we will dip these patterns. Building Zebra Design and proving the sprints may mean working harder in the establishment phase. That progression is accepted.
What we absolutely avoid: the Deep Work trap. Deep Work over-optimised working structure to what the team wanted, became too expensive, couldn't adapt when the market shifted, and the company died. Everyone suffered — finding new work, new clients, huge stress. The cautionary tale is real. The balance is: protect health enough to sustain, push enough to establish. Not either extreme.
The Diagnosis (5 Perspectives, 1 Conclusion)
Five independent analyses — optimistic, pessimistic, neutral, Claude synthesis, and Gemini Pro 3 — all converge: the collapse had multiple causes, not one. Each cause has a specific fix. The initial instinct (extend sprints to 2 weeks) treats the symptom. The root causes are below.
Law 9 (Failure): This diagnosis is irreducibly complex. Simplifying to "just take more time" would break it.
| What Broke | Root Cause | Fix |
|---|---|---|
| 2am finishes | Build hours exceeded 4-5hr/day limit | 4-hour hard cap. Stop at 15:00. |
| Exercise dropped | Work competed with exercise for morning time | Personal trainer every sprint morning. Outsourced executive function. |
| No regulation before work | Sprint pressure skipped the morning anchor | Foncil morning stack: sea swim + PT + transition ritual, then work. |
| Worst work at worst time | Complex components attempted when energy was lowest | Front-load complexity. Architecture on Day 2. Polish on Day 4. |
| Cumulative depletion | 3 consecutive sprints with no real recovery | Recovery rule (already locked in) + async gap = mandatory rest. |
| Scope exceeded capacity | Delivered 25+ screens when fewer would have sufficed | Technical freeze by Thursday noon. If it's not working, cut it. |
Law 1 (Reduce): Six fixes. All free except €150-300/sprint for a personal trainer. Zero changes to the external pitch.
Why Not 2 Weeks
The 2-week timeline extension was explored thoroughly and rejected as a first move:
- ADHD activation requires deadline pressure. "Aim for 1 week, budget for 2" is ambiguity — the worst of both neurologies. Autism needs clear structure, ADHD needs clear urgency. A flexible timeline provides neither. Every analyst flagged this.
- Parkinson's Law. Work expands to fill time. Extra days become extra polish, not extra rest. The AuDHD brain over-delivers when given room.
- "One week. One flow. Shipped." is the sharpest sentence in the business. It compresses buying decisions, communicates capability, separates from every competitor. The 12-18 month first-mover window depends partly on this speed claim.
- The collapse was consecutive, not structural. Sprint 1 alone might have been fine. The breakdown came at Sprint 3 of 3. The fix is recovery rules between sprints, not longer sprints.
Law 10 (The One): Subtract the timeline change. Add the interventions that address what actually broke.
The 2-week structure becomes the fallback — but only if the Full Stack still produces energy collapse after 2-3 sprints. And if it's adopted: commit fully to 2 weeks with distinct Week 1 / Week 2 phases. Never "aim for 1, budget for 2."
The Solution: Full Stack Sprint Protection
Keep the 1-week sprint. Fix what actually broke.
A sprint day follows the same anchor sequence that already works with a newborn and AuDHD — no fixed times:
Wake when family wakes (no alarm, no set time)
→ Get outside. Sunlight.
→ Family time — Coco and Bene, as long as it needs
→ Exercise / workout — takes as long as it takes
→ ONLY after exercise completes: work begins
→ Deep work (4 hours maximum from start)
→ Done.
Some days work starts at 11am. Some days at 4pm. Both are fine. The family time that a "hard stop" was designed to protect has already happened — in the morning. Working until done in the evening is OK because the sequence, not the clock, protects what matters.
The only sprint-week adaptation: Personal trainer booked at ~10-11am. This creates external accountability for exercise without breaking the flexible morning. Family time still comes first. PT ensures exercise actually happens under sprint pressure — the specific thing that failed at Research Tech.
Law 3 (Time): 4 hours of deep work at 12-20x AI productivity = what a traditional designer delivers in 3-6 days. The quality stays at 9/10. The organism survives. The clock is irrelevant — the sequence is everything.
The rules (non-negotiable during sprint weeks):
- 4-hour daily cap. 2am finishes are a scope failure, not a time failure.
- Personal trainer every sprint morning. Someone is waiting. Can't skip. €150-300/sprint week.
- Front-load complexity. Hardest work on Day 2 when energy is highest. Polish on Day 4.
- Technical freeze by Thursday noon. Not working by then? Cut it. No Friday debugging panic.
- Async gap = mandatory rest. No Design Canvas. No admin. No outreach. Just recovery.
- Cap consecutive build sprints at 2. Never repeat the 3-in-a-row that broke Research Tech.
Law 6 (Context): The personal trainer is not a luxury. It is the cheapest executive function outsourcing available — €150-300 to make exercise structurally uncancellable.
The Decision Sequence
- Sprint 4 (next client): Full Stack active, 1-week structure. Track: daily hours, exercise (yes/no), energy at sprint end (1-10), recovery time needed.
- Sprint 5: Same. Compare data.
- If sustainable: Lock in. Raise prices. Build toward fewer sprints at higher rates.
- If still collapsing: Switch to committed 2-week structure — distinct Week 1/Week 2 phases, clear internal milestones, no ambiguity.
The Reframe (From Gemini Pro 3)
At €15k per sprint, the annual target requires 6 sprints. That's 6 weeks of delivery in a 52-week year. The business model already has massive built-in recovery — the problem is recognising it and not filling it with anxiety.
Law 2 (Organise): 6 weeks of intensity. 46 weeks of life. The many (recovery weeks) appear as one (the gap between sprints).
Full analysis: workings/energy-collapse-sprint-timeline/ — 7 documents, 5 perspectives, 20+ alternative solutions explored.
Travel and Projects Cannot Overlap
Too much pressure. Research Tech had travel days conflicting with build schedules. This can't happen again. Sprint weeks are protected from travel.
Timeline Adjustments Are Real
Life with a newborn and AuDHD meant moving days and extending over a week. This happened and will happen again. The async gap in the sprint model absorbs some of this, but the build days need to account for life.
Only the Founder Had Project Context
Daniel (team member) turned up to workshops without full context of scope boundaries and the value being delivered. This led to small scope creeps — extra user research sessions, additional questions, additions that were hard to push back on because expectations hadn't been set upfront.
Solution: Onboard all team members through the project/onboarding flow before Workshop 1. Anyone attending a workshop needs to understand: what the sprint delivers, what it doesn't, and where the boundaries are. This reduces cognitive load in Workshop 1 and prevents mid-sprint scope conversations.
Facilitator Directness Needs Calibrating
Bartek: "You never say something negative. Sometimes it's hard to tell if the idea is stupid." Technical founders want blunt assessments. Calibrate directness to the client.
€5k Was Too Cheap
Below even the pessimistic floor valuation. Never go below €8k again.
Decisions to Lock In
| Decision | Evidence | Answer |
|---|---|---|
| Day 3 scope lock as standard | Worked exceptionally well — better than Deep Work's approach | Yes — standardise as methodology |
| Shortened workshop (2.5hrs) | Sketching proved obsolete. Journey refinement + scope lock is what matters. | Yes — test next sprint |
| Prototype-first feedback as principle | Working code workshops confirmed more productive across all 3 sprints | Yes — document as Zebra principle |
| Validation as separate bookable sprint | Testing was highest-value part but crammed into gaps. Deserves its own sprint. | Yes |
| Supervote for feature prioritisation | Quick, efficient, no debate. Worked as designed. | Yes — standard |
| Mandatory recovery rule | Energy collapse proves this isn't optional | Yes — after 2 sprint weeks: 1 week blocked. After 3-sprint transform: 2 weeks blocked. |
| €8k price floor | €5k was below pessimistic valuation. Szymon: "a bargain." | Yes — never below €8k |
| Sprint Retrospective Call (1 week post-delivery) | Testimonials, feedback, upsell — currently happening ad hoc | Yes — standardise |
| React stack as default | Worked well, reusable, familiar | Yes |
| Onboard all workshop attendees | Scope creep from uninformed team members | Yes — build into pre-sprint prep |
| Sprint structure: Build + Build + Validate | Testing deserves dedicated sprint, not async gaps | Yes — update offering |
| Full Stack sprint protection (4hr cap, PT, front-load, Thursday freeze) | 5 perspectives unanimously reject 2-week extension as first move. Root causes are hours, exercise, and consecutive sprints — not timeline. | Yes — implement Sprint 4. Track data. 2-week is fallback only. |
| Personal trainer during sprint weeks | Exercise is neurological medicine. PT = outsourced executive function. €150-300/sprint. | Yes — book PT when client books sprint. |
| Cap consecutive build sprints at 2 | Energy collapse happened at Sprint 3 of 3. Standalone sprints may be sustainable. | Yes — never 3+ consecutive build sprints. |
| Technical freeze by Thursday noon | 2am finishes were likely code/debugging, not design. Cut scope, don't extend time. | Yes — test Sprint 4. |
9 February 2026
Source: All files in 2-sprints/post-research-tech-strategy-to-do/, final presentation transcript, business plan, today.md