# Developers in the age of AI: what decides who gets the contract

**Canonical address:** https://optimistra.com/en/blog/developers-in-the-age-of-ai
**Author:** `michel-wozniak`, https://optimistra.com/team/michel-wozniak
**Category:** `What NLP Makes Possible`
**Published:** 2026-11-18
**Updated:** 2026-11-18
**Reading time:** 11 min, 2296 words

## In brief

AI now resolves almost all the tickets in a benchmark test, and the market for developers has tightened. What decides who gets the contract increasingly lies in the relationship with the client and in the state you are in under pressure. These human skills can be trained, and Neuro-Linguistic Programming (NLP) works on them through three pillars.

---

At the end of September 2025, on a benchmark made of real GitHub tickets, the best artificial intelligence (AI) model resolved nearly 70% of them. In September 2026, it resolved 97%, and the organization that runs the test stopped running it, because the test is saturated. The developer’s job has changed more quickly in one year than in the previous ten.

I was a developer, then an integrator and project manager, for 13 years in the automotive industry, and I have worked in AI for more than 20 years. What I see today in the developers we train comes down to one sentence: what decides who gets the contract, or whom the company keeps, is increasingly the relationship with the client, and the state you’re in when the pressure rises. That criterion isn’t written in any specification.

[Neuro-Linguistic Programming (NLP)](/en/blog/what-is-nlp) trains these skills, and it is what Nathalie de Marcé and I have taught together for about a decade. Here is the path, in six steps:

1. what has changed for developers in one year;
2. what clients and employers now look for;
3. why it all comes down to details you can’t see in yourself;
4. a few questions to ask yourself today;
5. NLP, as a developer sees it;
6. where to start with us.

## 1. What has changed for developers in one year

The test I’m referring to, SWE-bench Verified, gathers real tickets taken from GitHub projects. Vals AI, which runs it on every model with the same method, measured nearly 70% of tickets resolved at the end of September 2025, then 97% in September 2026, before dropping it. Usage followed. At METR, an independent institute that evaluates these models, the use of AI was closely supervised in late 2025; by spring 2026, many of its engineers were letting it carry out on its own tasks that would have taken them hours.

The market followed too. On freelancermap, a large freelance platform for the German-speaking region, Switzerland included, each published project receives 11 applications on average. In Switzerland, among application programmers, software developers, and systems analysts, unemployment rose almost twice as much as in occupations with little exposure to AI, according to the KOF Swiss Economic Institute at ETH Zurich. In France, the professional and managerial job postings for developers published on the site of APEC (Association pour l’emploi des cadres) fell from 31,330 in 2022 to 12,690 in 2025. These are postings, not hires, but the trend is clear.

So each contract matters more. On Malt, an experienced freelance developer lists an average of €576 a day in France, and a development project given to a freelancer lasts a median of 9 months in the German-speaking region (freelancermap). A contract that goes elsewhere means months of billing going to someone else. For an employee, the stakes are the same, whether in a job interview or on the day the company has to choose whom to keep.

## 2. What clients and employers now look for

In Switzerland, 66% of employers would be willing to pay more for communication and teamwork skills (ManpowerGroup Switzerland, press release of August 11, 2026, on the third-quarter barometer). In its 2025 report on the future of jobs, the World Economic Forum (WEF) asked more than 1,000 large employers which skills they consider essential today. Programming is cited by 17% of them, empathy and active listening by 50%.

Freelance platforms show it another way. Researchers analyzed about 11.66 million transactions on Fiverr, then ran controlled experiments (Hart et al., 2026). Even a delay of 5 to 10 minutes in replying to a client clearly lowers the chances of being hired, because speed signals that you will be responsive afterward. When the reply seems automatic or written by an AI, that advantage disappears.

The reason is that trust comes before competence. Face to face with someone, we judge their warmth before their competence, and warmth carries more weight (Fiske, Cuddy, and Glick, 2007). In three organizations studied, two of them IT companies, when people disliked a colleague, that colleague’s competence hardly mattered when they chose whom to work with (Casciaro and Lobo, 2008). AI strengthens this need for trust. In the United States, among employees who received from colleagues AI-generated work that looked polished but was empty, 42% then judged its author less trustworthy (BetterUp Labs and Stanford Social Media Lab, 2025).

Matt Pocock, a developer who trains developers, says the same about the people he teaches, in an interview with Nate Herk (2026). In his view, the people who really succeed are the ones who communicate best. That has always been true, for developers as for everyone else, and every top-tier developer he has seen was an excellent communicator.

He takes this one step further for work with AI agents: the quality of your thinking and the quality of your language are decisive, not only when you talk to the client, but also when you talk to the agent. [Watch the interview](https://www.youtube.com/watch?v=Z8lf5oWKfj0).

Take a moment to imagine what comes next. Your résumé is in the hands of a recruiter, or of a client choosing between two providers. Under your experience, that person reads these lines: analytical thinking, resilience and agility, leadership and social influence, curiosity and lifelong learning. Just above is the course that developed them, with a certificate to back it up.

Your reader stops at one of them and asks you for an example. You have one, a specific one, and you tell it in a steady voice, looking them in the eye.

*How do you feel, at that moment? And compared with those whose résumé says nothing of this, how much do you think your chances improve?*

## 3. Why it all comes down to details you can’t see in yourself

We don’t see ourselves, and research measures this even among developers. In early 2025, METR ran a controlled trial with 16 experienced open-source developers: they believed they were 20% faster with AI, and the stopwatch said they needed 19% more time. In front of a client, it’s the same: a word of jargon, a defensive answer, and you don’t notice.

Our states pass to others, even when we hide them. In a Stanford study, when one of two people discussing an upsetting subject suppressed their emotions, the other person’s blood pressure rose, and the connection formed less well (Butler et al., 2003). The client doesn’t read your code; the client reads your state.

Finally, stress cuts off access to what you know how to do. Under stress you can’t control, the part of the brain that thinks things through is the first to drop out (Arnsten, 2009). The skills are still there; you simply can’t reach them. That’s why someone can be brilliant in a code review and awkward in front of a client. And orchestrating several AI agents in parallel means switching constantly from one task to another, which draws on the prefrontal cortex, the region of the brain most sensitive to stress; a meta-analysis shows that after acute stress, cognitive flexibility drops (Shields, Sazma, and Yonelinas, 2016). No study yet covers developers who run agents: the conclusion I draw, that a practical ability to manage your own stress becomes essential for orchestrating them, is my own, and I offer it as exactly that.

That’s why I always come back to the same idea. A developer (like anyone who deals with other people) can delegate almost everything. But there are two things they cannot delegate: managing their own inner state, and the impact their words will have on others – and on themselves. An AI can write the code. You’re the one who then walks into the room, or onto the call, where it will be presented.

## 4. The questions to ask yourself today

These questions are a way to check what is at play for you, today. Take the time to answer them honestly, in writing if you can.

- On your last proposal, how many competitors did you imagine you were up against?
- In your last meeting with a client, how long did you talk about your stack, and how long about what was worrying them?
- The last time a client criticized your work, what was your first sentence?
- Before a demo that mattered, what state were you in when you walked in, and had you chosen it?
- When an agent hands you convincing code, what do you rely on to judge that it’s correct?
- What can a client get only from you?

If several of these questions gave you pause, you know where to put your effort. The good news is that all of this can be trained.

## 5. NLP, as a developer sees it

The name speaks to a developer. Neuro-Linguistic Programming looks at the programs each of us runs without thinking, in the way we communicate, react, and decide. It began in California in the 1970s, when its co-creators, Richard Bandler and John Grinder, observed how experts known for their results went about their work, in order to make their way of doing things teachable.

We teach it through three pillars, each feeding the others:

- **Communication**, with other people, with yourself, and with an AI: bringing out the need the client doesn’t state, talking about your work without jargon, recognizing the language of influence, and phrasing a request to an AI better, since a good prompt is communication too.
- **Inner states**: regaining your calm before a meeting that matters, bringing tension down in the moment, taking back control when impostor syndrome shows up.
- **Strategies**: spotting your automatic patterns, those that serve you and those that hold you back, and installing others.

An automatic pattern doesn’t change in one reading. A meta-analysis of 20 studies shows that a new habit generally takes 2 to 5 months to settle in, far from the 21 days so often quoted (Singh et al., 2024). Like a system, the brain is configured through use: with practice, the new program overrides the old one, and it’s the new one that runs. I wrote about this at greater length in [how long does it really take to form a habit?](/en/blog/how-long-to-form-a-habit)

I learned it the hard way. On September 7, 2011, after months of working day and night, I burned out. I still had my engineering skills, but my memory seemed switched off; a few hours earlier I had felt at my peak, and I hadn’t seen anything coming. It was while rebuilding myself, with Richard Bandler and with Tony Buzan, the inventor of Mind Mapping, that I discovered what no engineering school had taught me: managing my inner state, and the impact of words.

## 6. Where to start

Nathalie de Marcé and I are the world’s first two French-speaking Master Trainers named by Richard Bandler himself, and we teach with a team of trainers and assistants. Over several years, we had our course validated by more than 20 brain and health professionals (PhDs in neuroscience, biology, and biophysics, physicians, and psychologists), reworking it with them until it agreed with what they know.

### The NLP Practitioner course, 8 days in person

The NLP Practitioner course runs over 8 days, in Switzerland, in a group, and Nathalie and I lead it together. It is the same course for everyone, whatever their profession. You practice the three pillars all day, first on yourself, then with others, on the real situations each person brings: for a developer, the meeting with a client, the demo in front of senior management, criticism of a piece of work. Assistants who are NLP Master Practitioners move between the groups to supervise the exercises. Scheduled sessions are taught in French; the course is given in English on request.

- The certificate is signed by Richard Bandler himself.
- After the course, twelve live group sessions on Zoom, one hour a month for a year.
- A WhatsApp group, where the trainers and assistants stay for at least a year.
- The OptimiStra® Platform: practice apps, one for each skill worked on in the course, on your phone or your computer, with nothing to install, so you can keep working on what you learned.
- Admission starts with a 30-minute admission call on Zoom, to check together that the timing and the format suit you.
- Your company can pay for the course: we issue the quote or the invoice in its name.

Why 8 days in a room, when you’ve always learned online? Because what matters most happens in what a screen conveys worst: the micro-movements of the face, the breathing, what the voice does half a second before the word. That is what you learn to perceive in a client.

### When it isn’t the right time

The course asks for real commitment: you work on yourself, and at times that means questioning yourself. It is not therapy, and it does not replace therapy. If you’re going through a truly difficult period, this isn’t the right time, and we’ll be glad to welcome you later. We bring the method, the practice, and the support; the results come from practice, and everything is built around it.

If you’d like to look at what is at play in your meetings, with your clients or in your company, I suggest a 30-minute video call. I’ll tell you frankly whether the NLP Practitioner course can be useful to you. The details of the course are on the [NLP Practitioner course page](/en/nlp-practitioner).

---

## Further reading

Two related articles: [the skills of 2030, or what AI won’t do for you](/en/blog/future-skills-2030-and-nlp), and [how to remain the author of your decisions when you work with an AI](/en/blog/ai-and-decision-making).

And if you’d rather follow a thread than browse at random, [our complete path to understanding NLP](/en/nlp) puts our articles in order.

*Article designed and written by Michel Wozniak, and optimized by artificial intelligence.*

---

## Frequently asked questions

### Do developers still need people skills now that AI writes code?

More than before. AI now resolves most of the tickets in a benchmark test, and what separates two developers increasingly lies in the relationship with the client and in composure under pressure. In Switzerland, 66% of employers would be willing to pay more for communication and teamwork skills.

### I’m an employee, not a freelancer. Does this apply to me?

Yes. The same criteria apply in a job interview, and on the day a company has to choose whom to keep. Trust comes before competence: in organizations studied by researchers, two of them IT companies, when people disliked a colleague, that colleague’s competence hardly mattered when they chose whom to work with.

### What does NLP offer a developer?

Neuro-Linguistic Programming (NLP) works on the programs each of us runs without thinking. We teach it through three pillars: communication, to bring out what the client needs, talk without jargon, and phrase a request to an AI better; inner states, to regain your calm before a meeting that matters; strategies, to spot your automatic patterns and install others.

### How long does it take to change an automatic habit?

A meta-analysis of 20 studies finds that a new habit generally takes 2 to 5 months to settle in, far from the 21 days so often quoted. That is why the NLP Practitioner course doesn’t stop on the last day: it includes twelve live group sessions on Zoom, one hour a month for a year.

---

## Sources

- [Vals AI, SWE-bench Verified, même méthode pour tous les modèles](https://www.vals.ai/benchmarks/swebench)
- [METR (19 mai 2026), Frontier Risk Report](https://metr.org/blog/2026-05-19-frontier-risk-report/)
- [freelancermap (2026), Freelancer-Kompass 2026](https://www.freelancermap.de/files/survey/freelancer-kompass-2026.pdf)
- [freelancermap (24 septembre 2026), Stundensatz als Softwareentwickler](https://www.freelancermap.de/blog/stundensatz-softwareentwickler/)
- [KOF, École polytechnique fédérale de Zurich (octobre 2025), étude sur l'IA et le marché du travail suisse](https://ethz.ch/content/dam/ethz/special-interest/dual/kof-dam/documents/newsletter/KOF_Studie_KI_Schweizer_Arbeitsmarkt.pdf)
- [APEC (février 2026), Les métiers cadres porteurs, édition 2026 (in French)](https://corporate.apec.fr/files/live/sites/corporate/files/Nos%20etudes/PDF/Les%20metiers%20cadres%20porteurs%20edition%202026.pdf)
- [Malt, Baromètre des tarifs, page Tech, relevé du 9 octobre 2026 (in French)](https://www.malt.fr/t/barometre-tarifs/tech)
- [ManpowerGroup Suisse (2026), Baromètre des perspectives d'emploi en Suisse, troisième trimestre 2026 (in French)](https://www.manpowergroup.ch/fr/meos)
- [World Economic Forum (janvier 2025), The Future of Jobs Report 2025](https://reports.weforum.org/docs/WEF_Future_of_Jobs_Report_2025.pdf)
- [Hart, VanEpps, Sezer et Amir (2026), Speed Is a Signal: When Faster Replies Increase Hiring Likelihood, Management Science](https://doi.org/10.1287/mnsc.2024.06185)
- [Fiske, Cuddy et Glick (2007), Universal dimensions of social cognition: warmth and competence, Trends in Cognitive Sciences](https://doi.org/10.1016/j.tics.2006.11.005)
- [Casciaro et Lobo (2008), When Competence Is Irrelevant: The Role of Interpersonal Affect in Task-Related Ties, Administrative Science Quarterly](https://doi.org/10.2189/asqu.53.4.655)
- [BetterUp Labs et Stanford Social Media Lab (septembre 2025), The hidden costs of workslop](https://www.betterup.com/blog/hidden-costs-workslop)
- [METR (10 juillet 2025), Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity](https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/)
- [Butler, Egloff, Wilhelm, Smith, Erickson et Gross (2003), The social consequences of expressive suppression, Emotion](https://doi.org/10.1037/1528-3542.3.1.48)
- [Arnsten (2009), Stress signalling pathways that impair prefrontal cortex structure and function, Nature Reviews Neuroscience](https://doi.org/10.1038/nrn2648)
- [Shields, Sazma et Yonelinas (2016), The effects of acute stress on core executive functions: A meta-analysis and comparison with cortisol, Neuroscience & Biobehavioral Reviews](https://doi.org/10.1016/j.neubiorev.2016.06.038)
- [Singh, Murphy, Maher et Smith (2024), Time to Form a Habit: A Systematic Review and Meta-Analysis of Health Behaviour Habit Formation and Its Determinants, Healthcare](https://doi.org/10.3390/healthcare12232488)
- [Nate Herk (2026), Matt Pocock: The Most Valuable Skill to Learn in the Age of AI, entretien vidéo, YouTube](https://www.youtube.com/watch?v=Z8lf5oWKfj0)
