AI in BC development · 2026 survey

Moving Mountains

How AI is reshaping the AL/BC landscape

Over the summer I talked to BC teams at more than a dozen companies across Europe. Some had agents writing most of the code. Others used no AI at all. So I asked: where are BC development teams with AI, really?

97 people who work with Business Central answered. This is the map: what they use, where work piles up, and where they expect to be in 12 months. One statement splits them almost evenly: "Within 12 months, our team will routinely put AI-generated AL code into production without a human reading every change." 49 % agree, 47 % don't.

Jan SchefflerFounder of PEQY, AI Code Reviews for AL/BC

The code moved. Everything around it is next.

A year ago, 7 % of respondents said AI wrote or substantially rewrote more than half of their team's new AL code. Today, 53 % do. The answers show what moves next: where work waits, what teams trust, how work is priced, and who builds the tools.

  1. 1 · The code moved53 %of respondents say AI writes or substantially rewrites more than half of their team's new AL code
  2. 2 · The bottleneck moved42 → 14 %of respondents name writing code as the main pile-up: a year ago (recalled), today
  3. 3 · The risk moved49 %of respondents expect AI code in production within 12 months without a human reading every change
  4. 4 · The money may move38 %of respondents expect fixed-price or value-based pricing to replace hourly billing for custom work within 12 months
  5. 5 · Teams build their own tools60 %of respondents expect to build their own agents or MCP servers within 12 months

Each move opens with a view from the survey's one open question: “What is the one thing about AI in BC development that you think people are getting wrong?” I grouped the answers into six themes, summarized in my own words rather than quoted, and set each one next to the numbers.

Start with the code itself: how much of it AI writes today, and how fast that changed.

Move 1 of 5

The code moved.

53 %

of respondents say AI writes or substantially rewrites more than half of their team's new AL code today. A year ago: 7 %. In 12 months: 77 % expect it.

Read it left to right. The red bands are the teams where AI writes more than half of the new AL code. The dashed stretch on the right is expectation, not fact.

Today, AI writes more than half of new AL code for 53 % of respondents, up from 7 %.What share of your team's new AL code is written or substantially rewritten by AI?
  • 76–100 %
  • 51–75 %
  • 25–50 %
  • 1–24 %
  • None

The totals hide the movement. Pick an answer in any column to follow the people who gave it into the next column.

A year ago, today, in 12 monthsWhat share of your team's new AL code is written or substantially rewritten by AI?

Tap an answer in any column to follow the people who gave it into the next column.

Today, 28 % answered “25-50%”. The red bands show their answers 12 months ago.

Among respondents in BC teams of up to five people, 73 % say AI writes more than half of their new AL code. In larger teams, 49 % do.

68 % say their organization encourages AI use in development, with tooling and budget.

If AI writes the code, writing is no longer where the work waits. So where does it wait now?

Move 2 of 5

The bottleneck moved.

42 → 14 %

said work piled up most at writing code, a year ago and today. Planning leads now (30 %). For next year, customer acceptance is the stage named most (23 %), and 24 % don't know yet.

The same view for the bottleneck: follow the people who named “writing code” a year ago to where they see the pile-up today.

A year ago, today, in 12 monthsWhere does work pile up the most in your team?

Tap an answer in any column to follow the people who gave it into the next column.

Today, 30 % answered “Requirements and planning”. The red bands show their answers 12 months ago.

Where AI writes more than half of the new AL code, 10 % still say work piles up most at writing code. Where it writes half or less, 23 % do.

Respondents use AI across the whole flow: 95 % use it to write AL code, and most also use it for documentation (76 %), requirements (67 %) and code review (67 %).

Where does your team use AI regularly today?
  • Writing AL code95 %
  • Documentation76 %
  • Requirements, estimation, specs67 %
  • Code review67 %
  • Writing tests63 %
  • Debugging and support tickets54 %
  • Building AI features into BC for customers (Copilot, agents)37 %
  • Upgrades and C/AL to AL migration35 %
  • Functional setup and configuration work29 %

Faster code raises the next question: who checks it before it ships?

Move 3 of 5

The risk moved.

49 %

expect AI code to reach production without a human reading every change, within 12 months.

"Within 12 months, our team will routinely put AI-generated AL code into production without a human reading every change."
Agree 49 %47 % Disagree
AgreeSomewhat agreeSomewhat disagreeDisagree

Teams that expect to ship AI code without a human reading every change also use more AI: it writes more than half of the new AL code for 74 % of them, and for 37 % of the others.

How do AL changes get to production?
No version controlDirect commitsPRs, human reviewPRs, CI compilesPRs, CI testsDon't know

Where pull requests run automated tests in CI, 71 % expect to ship AI code without a human reading every change. Where pull requests get human review only, 37 % do.

If code gets cheaper to write, the way BC work is sold has to answer for it.

Move 4 of 5

The money may move.

63 %

expect their customers to want faster delivery or lower prices because of AI, within 12 months. 55 % expect customers to ask for AI and agent features inside BC.

How will your customers' expectations change within 12 months?More than one answer possible. Pick one.

63 %

Each dot is one of the 97 people who answered.

Inside their own organizations, respondents expect the business model to move. 38 % expect fixed-price or value-based pricing to replace hourly billing for custom work. 45 % expect fewer developers to be needed for the same output.

Which of these do you expect in your organisation within 12 months?
  • Fixed-price or value-based pricing replaces hourly billing for custom work38 %
  • Fewer developers needed for the same output45 %
  • Consultants build customisations without developers21 %
  • Headcount grows because of AI-driven demand11 %

Where AI already writes more than half of the team's new AL code, 57 % expect fixed-price or value-based pricing to replace hourly billing for custom work. Where it writes half or less: 18 %.

Both views are in the data. 8 of 31 open answers say AI will not replace the BC developer. 45 % expect fewer developers for the same output, and 11 % expect headcount to grow because of AI-driven demand.

Expectations like these need tooling. Teams are not waiting for someone to sell it to them.

Move 5 of 5

Teams build their own tools.

60 %

expect to build their own agents or MCP (Model Context Protocol) servers within 12 months. 26 % expect to buy commercial AI development tooling.

Many already have: 29 % use agents or MCP servers they built themselves, next to GitHub Copilot in VS Code (71 %), Claude Code (62 %) and Microsoft's AL agents (43 %).

Which tools does your team use for BC development?
  • GitHub Copilot in VS Code71 %
  • GitHub Copilot CLI or agent mode25 %
  • Claude Code62 %
  • Cursor, Windsurf or similar8 %
  • Microsoft AL agents, AL MCP43 %
  • ChatGPT, Claude or Gemini chat, ad hoc39 %
  • Agents or MCP servers we built ourselves29 %
  • Other AI development tools12 %
  • None1 %
  • Don't know0 %

Teams that already use agents or MCP servers they built themselves also run automated tests in CI more often: 50 % of them, and 10 % of the others.

For the tooling they will rely on most, teams still look to Microsoft (45 %) and general-purpose tools (37 %).

Which AI development tooling do you expect to rely on most in 12 months?
  • Microsoft-native: GitHub Copilot and AL agents45 %
  • General-purpose tools such as Claude Code or Cursor37 %
  • BC-specialised third-party tools4 %
  • Self-built8 %
  • We do not expect to use AI development tooling1 %
  • Don't know4 %

Last: what respondents want to fix in the next 12 months, and what stands in the way.

What it takes

What teams want to fix next.

30 %

would most like to improve the time from change to production in the next 12 months. Next: bugs that reach customers (24 %).

Which of these would matter most to improve in the next 12 months?
  • Time from change to production30 %
  • Bugs that reach customers24 %
  • Time changes wait for review7 %
  • Effort per BC upgrade4 %
  • Support ticket load13 %
  • Share of billable time9 %
  • Nothing in particular8 %
  • Don't know4 %

The blocker named most is output quality: invented objects and APIs (43 %). Then skills and time (36 %) and cost (35 %).

What's holding back more AI in your BC development today?
  • Output quality, invented objects and APIs43 %
  • Senior reviewers are not comfortable relying on AI27 %
  • Data residency and compliance20 %
  • Customer contracts8 %
  • Cost and licensing35 %
  • Skills and time36 %
  • Tools don't understand AL/BC well enough25 %
  • Nothing significant16 %
  • Another blocker5 %
  • Don't know4 %

Five moves, one direction. The code moved first: for 53 % of respondents, AI now writes more than half of their team's new AL code. The bottleneck followed it out of the editor: into planning today and, respondents expect, into customer acceptance next year. Trust is split: they divide almost evenly on shipping AI code without a human reading every change. The money and the tools are next, and 60 % expect to build at least part of their own. What ties the five together: every move pushes the hard part of BC development away from writing code and towards checking, deciding and delivering it.

Who is on this map

When a new development tool comes out, 43 % of respondents say their team tries it early, before others do. 34 % wait until peers they trust use it. Read this map as what these 97 respondents say, not as the whole BC community.

Thank you to the 97 people who took the time to answer.

Thanks to Aziz Skhiri, Alex Silva, James Muguro and Isaac Ampaire for writing their own posts about the survey, to Harry Cruz, Miljan Milosavljevic, Bert Verbeek, Danilo Zivanovic, Jan Schnitker, Andreas Reger, Matthias Küblböck, Airi Ifinedo, Edward Mihai Ionescu, Fernando Artigas Alfonso and Søren Friis Alexandersen for sharing it, and to everyone who commented and nudged it along.

If you want to talk about AL/BC code review, message me on LinkedIn or see how PEQY works.