AI-First Product Development

You have 50 engineers and you ship slower than when you had 5.

The bottleneck isn't your people or your tools. It's the process around them: specs, approvals, handoffs, and gates designed for a world where building was expensive. AI made building cheap. Your process hasn't caught up.

Book a call

The Problem

Your product development process was designed when a wrong bet cost six months. Heavy upfront planning made sense. So organisations built insurance: detailed specs before anyone writes code, architecture reviews before anyone picks a framework, approval gates before anything goes live.

AI collapsed the cost of building. Testing an idea takes hours, not months. But most organisations still pay the old insurance premium: planning for certainty in a world where they could just try it and see.

The result: business waits for engineering to show what's possible. Engineering waits for business to define requirements. Design waits for both. Nothing moves. The cost isn't just engineering time. It's market opportunities that expire while your process catches up. Bolting AI tools onto this broken pipeline (Copilot for developers still waiting 8 weeks for approved specs) just makes one step faster in a system where every other step is the bottleneck. When those tools hit a tangled codebase with hidden dependencies, they generate confident garbage faster.

The Research

Well-documented findings from decades of software engineering research.

15–25% flow efficiency

Most software teams. For every 10-day feature, 7–8 days are spent in queues, waiting for reviews, approvals, or the next team in line.

— Kersten, Project to Product (2018)

1/6th of time is coding

Only ~16% of a project schedule is actual coding. The rest is planning, testing, and waiting. This was measured in 1975. It hasn't improved.

— Brooks, The Mythical Man-Month (1975)

Approval boards make things worse

External approval processes like change advisory boards and architecture committees hurt both delivery speed and system stability. Larger batches, longer queues, higher risk.

— Forsgren, Humble & Kim, Accelerate (2018); DORA Reports

AI without process change backfires

Higher AI adoption correlated with −1.5% throughput and −7.2% stability when process fundamentals weren't addressed. Faster coding into a broken pipeline is faster waste.

— DORA State of DevOps Report (2024)

A reference point

What good looks like

1 day

Idea → Experiment

Does this concept work at all?

1 week

Experiment → Prototype

End-to-end, proves the architecture.

1 month

Prototype → Production

Live, real users, real data.

6 weeks

Production → Outcome

Did it move the needle?

Most organisations we talk to are 5–10× slower at every stage. The gap between these reference points and your reality is the transformation opportunity.

Our approach

Three shifts that compound

Subscribe, don't prescribe

Give teams an outcome to achieve, not a specification to execute. AI makes exploration fast enough that planning upfront costs more than trying it.

Everyone becomes a maker

Collapse the gap between the people who understand the problem and the people who can build the solution. When a product owner can prototype, the handoff disappears.

Build to learn, don't plan to build

Instead of specifying, reviewing, and approving, just build. A constrained prototype in weeks replaces specs, architecture docs, and planning decks with running code.

What this looks like

From specs to shipping

A large enterprise ran product development the way most do: detailed upfront planning, structured handoffs, Scrum teams executing against predefined scope. Thorough, but surprises showed up late.

We set up a small, AI-first team mixing product, engineering, and UX around a single commitment: build a constrained prototype that proves one end-to-end workflow. Product people involved daily, discovering edge cases, explaining business rules, subscribing to results rather than prescribing solutions.

The team shipped the target use case in one cycle and covered significantly more ground than planned. People who used to work on opposite sides of a handoff were building in the same repo. The results got us invited to present the approach to company leadership.

There was no going back.

How we start

Start small. Prove fast.

Every engagement produces working software. We learn your codebase by shipping a feature, not by writing an assessment. We set a baseline in the first engagement and measure progress each cycle: delivery speed, team autonomy, engineering quality. You see the numbers, not just the narrative.

Spark

1 day

Live building session with your team. Take a real problem from your backlog, prototype it together. You walk away with working code and a visceral sense of the gap.

Best if you want to see the gap before committing.

Embedded Week

1 week

One senior engineer, in your team, doing your work. Ships something real. Surfaces blockers by hitting them, not by interviewing about them.

Best if you know something's broken and want to find out what.

Pilot Sprint

2 weeks

One concrete initiative, run AI-first end-to-end. Working software in production. The assessment is a byproduct of the work.

Best if you're ready to run a real initiative differently.

These entry points convert into ongoing outcome cycles, 6-week engagements with agreed targets, measured against baseline. But the first step is always: ship something real.

Why metosin

Judgment, not just speed.

Everyone can vibe-code. Only seasoned engineers know when AI output is production-worthy and when it's dangerous. Our engineers average 20+ years building production systems. AI makes us faster. Experience makes us safe.

Engineers who own outcomes, not tickets.

Our people build production systems, read the room, find the sponsors, and make product decisions without a spec. One senior engineer replaces an entire spec → estimate → review → build pipeline.

High agency, not consulting.

We don't observe and report. We take ownership, ship working software, and pull your team along. We find the high-agency people inside your organisation and amplify them.

Proven in hard industries.

Telco. Adtech. Complex regulatory environments, real-time systems, massive data volumes. We make autonomous decisions because we understand the problem space, not just the tech stack.

Tell us what's slow.

The organisations that figure out AI-first product development now will have a structural advantage for years. Book a 30-minute call. We'll map your process and show you where the leverage is.

Book a call

AI-First Product Development

You have 50 engineers and you ship slower than when you had 5.

Book a call

The Problem

The Research

Well-documented findings from decades of software engineering research.

15–25% flow efficiency

Most software teams. For every 10-day feature, 7–8 days are spent in queues, waiting for reviews, approvals, or the next team in line.

— Kersten, Project to Product (2018)

1/6th of time is coding

Only ~16% of a project schedule is actual coding. The rest is planning, testing, and waiting. This was measured in 1975. It hasn't improved.

— Brooks, The Mythical Man-Month (1975)

Approval boards make things worse

External approval processes like change advisory boards and architecture committees hurt both delivery speed and system stability. Larger batches, longer queues, higher risk.

— Forsgren, Humble & Kim, Accelerate (2018); DORA Reports

AI without process change backfires

Higher AI adoption correlated with −1.5% throughput and −7.2% stability when process fundamentals weren't addressed. Faster coding into a broken pipeline is faster waste.

— DORA State of DevOps Report (2024)

A reference point

What good looks like

1 day

Idea → Experiment

Does this concept work at all?

1 week

Experiment → Prototype

End-to-end, proves the architecture.

1 month

Prototype → Production

Live, real users, real data.

6 weeks

Production → Outcome

Did it move the needle?

Most organisations we talk to are 5–10× slower at every stage. The gap between these reference points and your reality is the transformation opportunity.

Our approach

Three shifts that compound

Subscribe, don't prescribe

Give teams an outcome to achieve, not a specification to execute. AI makes exploration fast enough that planning upfront costs more than trying it.

Everyone becomes a maker

Collapse the gap between the people who understand the problem and the people who can build the solution. When a product owner can prototype, the handoff disappears.

Build to learn, don't plan to build

Instead of specifying, reviewing, and approving, just build. A constrained prototype in weeks replaces specs, architecture docs, and planning decks with running code.

What this looks like

From specs to shipping

A large enterprise ran product development the way most do: detailed upfront planning, structured handoffs, Scrum teams executing against predefined scope. Thorough, but surprises showed up late.

There was no going back.

How we start

Start small. Prove fast.

Spark

1 day

Live building session with your team. Take a real problem from your backlog, prototype it together. You walk away with working code and a visceral sense of the gap.

Best if you want to see the gap before committing.

Embedded Week

1 week

One senior engineer, in your team, doing your work. Ships something real. Surfaces blockers by hitting them, not by interviewing about them.

Best if you know something's broken and want to find out what.

Pilot Sprint

2 weeks

One concrete initiative, run AI-first end-to-end. Working software in production. The assessment is a byproduct of the work.

Best if you're ready to run a real initiative differently.

These entry points convert into ongoing outcome cycles, 6-week engagements with agreed targets, measured against baseline. But the first step is always: ship something real.

Why metosin

Judgment, not just speed.

Engineers who own outcomes, not tickets.

High agency, not consulting.

We don't observe and report. We take ownership, ship working software, and pull your team along. We find the high-agency people inside your organisation and amplify them.

Proven in hard industries.

Telco. Adtech. Complex regulatory environments, real-time systems, massive data volumes. We make autonomous decisions because we understand the problem space, not just the tech stack.

Tell us what's slow.

The organisations that figure out AI-first product development now will have a structural advantage for years. Book a 30-minute call. We'll map your process and show you where the leverage is.

Book a call