# When AI can do a bit of everything, what should new hires go deep on?

2026.10.14 · KaYu Ling · Business perspective · Hiring in the age of AI, part 3

Research, slide decks, programming: you can ask AI for all of it. Work that took people days can now take shape in a short time.

So what should companies have new employees learn? Until now, the norm has been to rotate them through a wide range of work first and let them find a speciality over several years. Does that way of developing people still make sense as it is?

Unlike AI built for a single job, a general-purpose LLM can help across a broad range of work: writing, research, analysis, code. That does not mean it is good at everything, but it is something like an all-rounder whose skills chart is an even pentagon. It is not obvious that people should aim for the same pentagon.

**What companies should develop is perhaps not people who can do everything to an average standard, but people who understand a particular job and workplace deeply, and can make decisions while checking AI’s answers.**

Earlier in this series I looked at [mutual understanding through long-term internships](https://www.meldotia.com/blog/ai-hiring-work-together) and [ways to give students and their AI accurate hiring information](https://www.meldotia.com/blog/ai-hiring-job-information). This time I want to look at what comes next: developing people after they join.

**What this piece proposes**

1. Have new hires do the basic work by hand first, then use AI to learn faster
2. Settle the first team, and use rotation to learn the whole company and bring that back
3. Let people move into new roles on ability, not years served
4. Turn internal hackathons and the idea box into paths to trial budgets and new roles

## If AI covers the breadth, people should build depth

Take someone developing new products for a retailer. Researching the market, sorting out competing products and drafting an early plan are all things AI can help with.

But “why don’t the customers at this shop buy?”, “what is giving the floor staff trouble?” and “what should we change this time?” sometimes cannot be settled with numbers and general principles. Some things only become clear once you see the shop floor, listen to customers and staff, and know the limits of budget and logistics.

Experts who understand the technology in depth are needed too. But not every employee has to master AI technology itself. Experts who understand their own field, whether healthcare, logistics, manufacturing, finance or retail, have a large part to play. The value lies in seeing that a suggestion from AI “sounds right in general, but won’t work here”.

This is not to say people need no knowledge outside their speciality. Without a shared baseline, it is hard to work with the team next door, or to judge whether a request to AI is the right one. **Keep the basics broad, and be clear about the areas to deepen over time.** What I want to change is how that time is divided.

## Do the basic work by hand first

Even so, I do not think new hires should hand all their work to AI from day one.

Will someone who has never done market research themselves notice the bias in what AI gathered? Will someone who has never read code themselves spot the risk in a change AI made? Taking only the results without understanding the process makes it hard to build the standards needed to judge them.

At first, have them find sources, question customers, organise data, form hypotheses, try and fail, on their own. Let them go through the whole process. This is not to keep them on inefficient work forever, but so they learn what good work looks like and where mistakes creep in.

After that, they should use AI freely. Redo the same task with AI and compare what got faster and what went wrong. Then spend the time freed from repetitive work on harder cases and on experience in the field.

| Stage | What the new hire does | What the company checks |
|---|---|---|
| Learn the basics by hand | Carries out the main steps themselves | Understands why each step is taken and what the quality standard is |
| Do the same work with AI | Uses AI and compares the result with their own | Notices errors and limits, and can fix them |
| Build expertise | Takes on hard problems in the field and decides | Can explain the reasons and own the outcome |

Do not settle the basics with a rule like “no AI for everyone for the first three months”. The steps and time needed differ by role and by how well each person understands. Check whether they have got it, then move on. In work where failure can cost lives or cause serious losses, of course, experience must be gained under proper supervision.

There is already evidence that AI can help beginners grow. A study published in The Quarterly Journal of Economics in 2025 followed 5,172 customer support agents and found that generative AI assistance raised issues resolved per hour by 15% on average. The least experienced improved the most, and the study also reported that it helped them learn. But it is one company and one kind of work, and the same effect will not hold for every job.[^1]

The lesson is neither “skip the basics” nor “AI gets in the way of learning”. The company’s job is **to provide, in the right order, both the experience of thinking for yourself and chances to learn faster with AI**.

## Settle the first team, then get to know the whole company

In [the first post in this series](https://www.meldotia.com/blog/ai-hiring-work-together) I proposed hiring where students and companies get to know each other through role-specific long-term internships, and the work done connects to the first placement after joining.

With that approach, development after joining changes too. The default is to settle the home team from the start and build expertise in that work.

That does not mean people can ignore other teams. A product planner needs to know how sales, manufacturing, logistics and customer support operate. Realistic decisions come from understanding how your choices affect other teams’ work.

So job rotation stays, but with a different purpose.

**Rotation is not for deciding again and again where someone belongs. It is for understanding the whole company and bringing that learning back to their first area of expertise.**

Combine whatever experiences are needed: short stints in other teams, shadowing in the field, cross-team projects. Afterwards, people return to their original team as a rule. But if someone tries other work and strongly feels “this is what I want to do properly”, check their wish and the receiving team’s conditions, and let them transfer formally.

That keeps a steady axis for building expertise without locking a person’s potential into their first placement. Rotation need not be the same length for everyone, either. Decide first what you want them to understand when they come back.

## If AI speeds up growth, do not measure development in years alone

AI can be someone to ask the same question many times, and someone to point out the weak spots in a plan. Used well, it may let new hires pick up knowledge and skills faster.

In Microsoft’s 2025 Work Trend Index, 83% of the leaders surveyed believed AI would let employees take on more complex and strategic work earlier in their careers. That is leaders’ outlook on the future, though, not a figure that directly proves people learn faster.[^2]

All the more reason for companies not to rely only on uniform development plans of the “in year N you get this far” kind. If a new hire understands the core work, uses AI well and can be trusted with customers and decisions, let them move on to the next thing. If the necessary skills are not there yet, provide the chance to learn regardless of seniority.

That does not mean everything should move to performance pay or pure meritocracy. Some work, like sales, has results that are easy to measure; some is long-term quality control or supporting the team. The chance to show results also varies with the projects, managers and team situation people happen to have.

What is needed is not ranking every employee on one number, but **opening up wide chances to show ability and take on new roles**.

## Turn internal hackathons into a path to new roles

One such chance is an internal hackathon.

Say a second-year employee in sales proposes a small service that fixes a problem they have seen many times with customers. Even with no development experience in their day job, they can team up with engineers and designers and build a prototype with AI. Understanding the customer’s problem is itself their strength.

Atlassian has run “ShipIt”, an internal event where people step away from their day-to-day work to build their own ideas, for many years. According to the company, the origins of Jira Service Management came out of it.[^3]

What we at Meldotia propose is not letting the hackathon end as a fun company event.

Give strong proposals a budget and time to prove themselves. Let them pull in members from other teams. If the trial goes well, promote it to a real business or improvement project, and open **a path for the person who proposed it to lead it, move to the work they want, or be given a bigger role regardless of seniority**.

Of course, giving a good presentation in a short event is not the same as running a business for years. So a new role should not be a verbal promise: first put the person in charge of the trial with real authority and budget, then expand the role based on results. Taking on new roles can become a way to discover abilities that ordinary performance reviews never showed.

It also matters that managers cannot hold on to their best people. Unless leadership and HR have the authority to approve these moves and the means to support the current team, the scheme will exist on paper only.

And so that people who can only join on the day are not at an advantage, allocate time from regular working hours. Set challenges that people outside technical roles can take on, and assess not just presentation skill but understanding of the problem, the results of the trial and contribution to the team.

## Make the idea box lead all the way to action

Hackathons alone cannot choose when ideas arrive. So companies also need an internal “idea box”, where anyone can propose an improvement or a new business at any time.

A plain submission form will eventually go unused. People cannot tell whether their proposal was read, how it was judged, or what it lacked.

So make the path from proposal to action clear.

- **Submit**: no finished business plan needed; “who is struggling with what” and a possible fix are enough.
- **Discuss**: people from other teams add views and their own experience, and recruit others to work on it if needed.
- **Test**: proposals that meet set criteria get a small budget and working hours to try it.
- **Build a business**: if results hold up, move it to a formal project with the proposer able to lead it.

Adobe has long run “Kickbox”, which lets employees test ideas with a small amount of money. It hands people a process and funding to validate an idea themselves, and was opened up to other companies in 2015. What is worth learning from is that it does not ask for a big business plan or an executive pitch at the start.[^4]

AI plays a big part here too. It can help with everything up to testing an idea: organising market information, working through hypotheses, building prototypes. **Not only people who are good at writing proposals, but people who know the problems on the ground, can more easily give shape to their ideas.**

But do not decide on votes alone. Popularity is not viability. Review the user’s problem, the impact, the cost and whether it can be done. Give reasons for proposals that are not taken forward, so they can be improved next time. Adjust who can see information that is hard to share internally.

Test problems posted to the idea box at a hackathon, and move the prototypes that come out of it into trials. Run them as one system, not as separate events.

## Build expertise, but do not lock people into their team

At first glance, these schemes seem to pull in opposite directions.

On one side, settle the first team and have people deepen their expertise in the same field. On the other, encourage challenges across teams and move strong employees into other work or new businesses.

But the two do not conflict. **Daily work needs a steady axis to build expertise. That does not mean a person’s potential has to be fixed by their first placement.**

What companies need is not only a plan for “what everyone should be able to do by year N”. It is to design, together, chances for new hires to learn the basics by hand, ways to learn faster with AI, experience making decisions in the field, chances to get to know other teams, and a system that can change someone’s role when unexpected talent appears.

You do not have to change much at first. In one role, define the standard for having learned the basics. Make clear the purpose of short rotations and which team people return to. Give the ideas chosen at an internal hackathon a budget and an owner, even a small one. Always reply to proposals sent to the idea box.

Then look not at how many hours of training people sat through, but at what they learned before they could be trusted with real work, how far proposals were carried out, and how people who were given new roles grew in them. I want to see how the time it takes to build expertise, and the chances to take on challenges inside the company, actually change.

**The more AI can fill in broad knowledge and tasks, the more clearly companies need to define the work people must understand deeply. And they need to let the people they have developed in depth move freely when it counts.**

Isn’t that how to develop people in the age of AI and grow the company at the same time?

*This post is a proposal informed by research and companies’ public information available as of 14 October 2026. How much AI shortens learning, the best way to develop specialists, and the business results of each scheme vary by industry and by how they are run. Translated from the Japanese original.*

## References

[^1]: Erik Brynjolfsson, Danielle Li, Lindsey Raymond, “[Generative AI at Work](https://academic.oup.com/qje/article/140/2/889/7990658),” The Quarterly Journal of Economics, 140(2), 889–942 (2025)
[^2]: [Microsoft, “2025: The year the Frontier Firm is born,” Work Trend Index](https://www.microsoft.com/en-us/worklab/work-trend-index/2025-the-year-the-frontier-firm-is-born) (23 April 2025)
[^3]: [Atlassian, “From Ideas to Impact: The Story of ShipIt”](https://www.atlassian.com/blog/development/from-ideas-to-impact-the-story-of-shipit) (14 January 2025)
[^4]: [Adobe, “Kickbox Open Sourcing Makes Innovation Easily Accessible for All Companies”](https://blog.adobe.com/en/publish/2015/02/09/kickbox-open-sourcing-makes-innovation-easily-accessible-for-all-companies) (9 February 2015)

## Series: Hiring in the age of AI

1. [When AI writes the application, hiring should move to working together](https://www.meldotia.com/blog/ai-hiring-work-together)
2. [When candidates search with AI, how should companies make themselves findable?](https://www.meldotia.com/blog/ai-hiring-job-information)
3. When AI can do a bit of everything, what should new hires go deep on? (this post)

## Contact

- Email: [contact@meldotia.com](mailto:contact@meldotia.com)
- Status: Open to collaboration
- https://x.com/kayuling
- https://www.linkedin.com/in/kayuling/
- https://www.threads.com/kayu.ling

HTML version: https://www.meldotia.com/