Plandek Thought Leadership
Key insights from the AI Transition Playbook, for turning AI spend into measurable engineering and business value.
The big question
Are you making every token count?
As AI usage scales, so does the spend. But more AI does not automatically mean more engineering productivity, or more business value.
That is tokenomics: the economics of AI use. The question is simple: are you getting enough useful engineering output and business value for what you spend?
But AI output and business output do not always rise together. This is the Agentic Value Gap: AI-generated output increases sharply, while the amount of value making it through the delivery system rises much more slowly.
The reason is simple: AI can accelerate an activity, without accelerating the system around it. Most organizations track AI adoption at one end and expect ROI at the other. The problem is that the value is won or lost in the middle – in how AI changes the way work flows through the rest of the delivery system.
Quiz
Are you really ready for agentic AI?
AI adoption is accelerating, and organizations have clear ambitions for what it should deliver. But the biggest barriers now sit in the engineering and organizational environment around the tools.
Common problem
The gap between rollout and AI ROI
ROI gets lost in the gaps between rollout and results, because AI can make one part of delivery faster, without making the whole system faster. In this example, coding is the main constraint before the introduction of AI. Once AI accelerates Build, the bottleneck moves downstream instead. More code is being produced, without value reaching customers.
When AI accelerates one stage, the slowest remaining stage becomes the new limit. Unless the rest of the delivery system can absorb the additional output, gains are lost in queues, rework, and delay – and the expected ROI never materializes.
What's changed
The operating model is changing too
Traditional software delivery often relies on sequential handoffs between product, engineering, testing and operations. The Agentic PSDLC moves toward an integrated Product and Software Development Life Cycle in which humans and AI work across one continuous system.
Choose the right level of autonomy based on the risk, reversibility and context of the work. Learn about the levels of autonomy, and how to choose them, in the AI Transition Playbook.
Solution
Five moves to from rollout to AI ROI
Meet the RACER™ Framework
AI amplifies the engineering system you already have. Strong foundations can accelerate the gains; weak processes, poor controls and hidden constraints can absorb them. Most organizations measure adoption and promise ROI. But the difficult work happens between those two stages.
To know whether AI is improving engineering performance, you first need to understand the system it is being introduced into. Plandek looks at engineering health across four dimensions:
If you cannot answer these before your AI rollout, it becomes very difficult to prove what AI changed afterwards.
Together with qualitative team feedback, these give you a baseline for Engineering Impact, so you can see what changes as AI adoption grows, and where gains are being lost.
Risks
What happens when agents scale faster than your controls?
AI agents can create work faster than your engineering system can absorb, review and govern it. Without the right context, controls and operating model, the upside gets diluted across six predictable pressure points.
Security and compliance
Agent-generated code can push vulnerabilities, data exposure and compliance failures into production faster.
Can you prove what your agents changed, and who approved it?
Quality and reliability
Defects and rework can accumulate faster than teams can catch them.
Are you shipping faster, or just moving defects downstream?
Technical debt
Poorly governed agents can amplify architectural inconsistency and technical debt at machine speed.
Is AI improving your architecture, or scaling its weaknesses?
People and review capacity
More agent output can overload reviewers and turn experienced engineers into the next bottleneck.
Have your senior engineers become the constraint AI was meant to remove?
Institutional knowledge
As agents do more work, critical engineering knowledge needs to move from people’s heads into usable context.
If key engineers stepped away, would your agents still understand the system?
Productivity & ROI
Token cost, rework, review effort and downstream constraints can absorb the gains from higher AI output.
How much of your AI-generated capacity is actually reaching the customer, and at what cost?
Prepare
Give your agents the right environment to work in
Together, harness and context engineering shape your tokenomics: whether AI spend is translating into useful engineering output.
Harness engineering
Is AI improving your architecture, or scaling its weaknesses?
It controls which models and tools they can use, what they can access and change, how work is orchestrated and evaluated, when humans intervene, and how activity is monitored.
Get the harness wrong, and AI can scale cost, defects and risk faster than it scales output.
Context engineering
Context engineering determines what the agent knows, and how efficiently it can work.
It’s a critical part of the harness. Give agents the right architecture, documentation, standards and business context, and they can make better decisions with fewer retries.
Get the context wrong, and you waste tokens on reprompting, correction and unnecessary model use, which increases AI cost without increasing useful output.
Measure
Can you prove your investment is paying off?
AI spend is rising fast. If you can’t show that engineering output is rising with it, you don’t know whether AI is creating value or simply adding cost.
The basic test is simple: compare the change in engineering output with the change in total engineering and AI cost.
Measure output against cost
Make every token count
See where AI is improving engineering performance, where gains are being lost, and where to focus next.
NEW
Contact us
UK Office
Unit 313 The Print Rooms, 164-180
Union St, London SE1 0LH
US Office
Floor 4, 1515 Mockingbird Ln,
Charlotte, NC 28209, USA