The expert is still in the room
Coding agents reward people who understand the problem. What would it take for your team to build, maintain, and improve its own solutions?
July 8, 2026
AI can make technical work accessible to more people. Knowing what good work looks like still matters enormously. This issue pairs research on domain expertise with a billion-dollar bet on engineers working alongside customers. We see an encouraging possibility: more people can build useful things, and the people who understand the problem have a valuable place at the keyboard.
Knowing the work helps you steer the agent
Anthropic’s analysis of roughly 400,000 Claude Code sessions associates greater task expertise with more successful sessions, including among people outside software occupations.
Our take: The colleague who can explain why a calculation is wrong has something a coding agent needs. Give that person a way to build and they may solve problems that never made it onto an engineering roadmap. These are session-level findings, so we’d resist the leap to “everyone can ship production software.” We would absolutely take them as a reason to widen who gets to try.
AWS is sending people along with the technology
AWS announced a $1 billion investment in forward deployed engineering, with embedded teams and a stated goal of leaving customers able to operate independently.
Our take: We like the exit condition: the customer should be able to keep going when the experts leave. Ask an AI partner what your team will understand, maintain, and change by the end of the work. A dazzling system that requires a permanent translator can become a very expensive dependency.
One useful move
Have a spreadsheet formula you understand but struggle to explain? Paste the formula and describe what it should calculate. Ask your usual AI tool:
Spreadsheet formula: {{spreadsheet_formula}} Cell context (labels, units, what each range means): {{cell_context}} Explain this formula in plain English, one step at a time. Then suggest a clearer version that calculates the same result. Tell me what assumptions you made about the cells.
Check the explanation against the sheet. Keep whichever version you can understand and maintain.
Something fun
Ask your chatbot to write a job reference for an apprentice wizard whose only reliable spell is making spreadsheets slightly narrower.
Give the reference three impressive accomplishments and one concern the referee is desperately trying to phrase positively. Keep the magic petty and the tone painfully professional.
“Occasionally turns a colleague into a pivot table” feels like a development opportunity.
Built in the past two weeks
We prepared the plan for “Just Use AI for That,” a workshop built around familiar HR tasks: turning email clutter into a decision brief, drafting manager communications, and pulling actions from meeting notes. The plan includes showing how to recover when a demo goes wrong, plus questions people can ask when leadership's entire brief is “use AI.” Useful practice needs room for the messy bit.