Developer AI Adoption Statistics 2026
By Axis Intelligence Research
Co-author: Sarah Mitchell | Last updated: September 15, 2026 | License: CC BY 4.0
Developer AI adoption reached 84% of surveyed developers and 90% of surveyed technology professionals in the most recent annual studies. According to Axis Intelligence Research, actual penetration across the thirteen stages of the software delivery lifecycle scores 19.6 out of 100, meaning the headline adoption figure overstates lifecycle integration by roughly 4.3 times.
Quick Answer
84% of developers use or plan to use AI tools, and 47.1% use them daily (Stack Overflow Developer Survey 2025, 49,000+ respondents, 177 countries). But adoption is not integration. According to Axis Intelligence Research, the Software Delivery AI Penetration score (SDAP) across all thirteen lifecycle stages is 19.6 out of 100, ranging from 39.3 for answer-seeking down to 6.0 for deployment and monitoring, where 64.8% of developers say they do not plan to use AI at all.
Key Findings
- According to Axis Intelligence Research, the lifecycle-wide Software Delivery AI Penetration score (SDAP) is 19.6 out of 100 as of the 2025 Stack Overflow field window, against a headline adoption rate of 84%.
- According to Axis Intelligence Research, build-stage tasks score an average SDAP of 22.2 while ship-and-govern tasks score 9.3, a 2.4-times penetration gap that no published developer survey currently reports.
- Google Cloud’s 2025 DORA report found AI adoption among 5,000 surveyed technology professionals reached 90%, with a median of two hours per day spent working with AI.
- The U.S. Bureau of Labor Statistics now projects 10% employment growth for software developers, quality assurance analysts and testers over 2025 to 2035, down from the 17.9% it projected for software developers over 2023 to 2033.
- TypeScript became the most used language on GitHub in August 2025 with 2,636,006 monthly contributors, a 66.63% year-over-year gain that GitHub attributes in part to typed languages making agent-assisted code safer to ship.
How Many Developers Actually Use AI in 2026?
Three independent annual studies now measure this, and they agree on direction while differing on level, which is exactly what you would expect from three different sampling frames.
| Study | Population | Sample | AI adoption | As of | Source |
|---|---|---|---|---|---|
| Stack Overflow Developer Survey 2025 | Developers and learners, 177 countries | 49,000+ | 84% use or plan to use | 2025 field window | Stack Overflow |
| JetBrains State of Developer Ecosystem 2025 | Developers, 194 countries | 24,534 | 85% regular use | April to June 2025 | JetBrains |
| DORA State of AI-assisted Software Development 2025 | Technology professionals | ~5,000 | 90% use at work | 2025 |
The gap between 84% and 90% is a population gap, not a contradiction. DORA surveys technology professionals inside organizations that already participate in DORA research, a self-selecting group with more mature delivery practices. Stack Overflow captures a wider public population including people learning to code, 19.5% of whom say they do not plan to use AI tools.
Daily use is the harder number. 47.1% of all Stack Overflow respondents use AI tools daily, rising to 50.6% of professional developers and 55.5% of early-career developers with one to five years of experience, and falling to 47.3% among developers with ten or more years. DORA reports a median of two hours per day spent working with AI, which is the single most useful intensity figure published anywhere in this space, because it converts a binary into a duration.
Sarah Mitchell: The adoption number stopped being interesting the moment it passed 80%. Everyone uses AI now in the same sense that everyone uses a search engine. What separates a team that has changed how it ships software from a team that has installed an autocomplete extension is not whether they use AI, it is which parts of the delivery pipeline they have handed over. That question has an answer in the data, and almost nobody has run it.
Which Parts of the Development Workflow Use AI? The SDAP Score
This is the section built on data that exists but has never been published in comparable form.
The Stack Overflow Developer Survey 2025 asked roughly 30,000 respondents to assign each of thirteen development tasks to one of five states: currently mostly AI, currently partially AI, plan to partially use AI, plan to mostly use AI, or do not plan to use AI for this task. Stack Overflow publishes the results normalized within each response option, which makes the five figures for any single task non-comparable. You cannot read off what share of developers use AI for testing.
Axis Intelligence Research reconstructed the all-respondent distribution. The method and its validation are in the Methodology section. The result is the first published table of AI penetration by lifecycle stage.
SDAP: Software Delivery AI Penetration
SDAP is an Axis Intelligence Research metric scoring how deeply AI has entered a given stage of the software delivery lifecycle, on a 0 to 100 scale.
SDAP = (share currently mostly AI × 1.0) + (share currently partially AI × 0.5)
Planned future use scores zero, because a plan is not a workflow. A stage where every developer had fully handed the work to AI would score 100. A stage nobody touches with AI scores 0.
| Rank | Lifecycle stage | SDAP | Mostly AI | Partially AI | Will not use AI | Source |
|---|---|---|---|---|---|---|
| 1 | Search for answers | 39.3 | 20.0% | 38.6% | 16.4% | Axis calculation on Stack Overflow 2025 |
| 2 | Learning new concepts or technologies | 28.8 | 12.3% | 33.0% | 27.2% | Axis calculation on Stack Overflow 2025 |
| 3 | Writing code | 26.5 | 6.2% | 40.6% | 24.0% | Axis calculation on Stack Overflow 2025 |
| 4 | Debugging or fixing code | 24.0 | 7.7% | 32.6% | 30.5% | Axis calculation on Stack Overflow 2025 |
| 5 | Generating content or synthetic data | 23.6 | 13.5% | 20.2% | 32.6% | Axis calculation on Stack Overflow 2025 |
| 6 | Documenting code | 22.1 | 11.5% | 21.1% | 32.4% | Axis calculation on Stack Overflow 2025 |
| 7 | Learning about a codebase | 19.2 | 7.8% | 22.9% | 33.3% | Axis calculation on Stack Overflow 2025 |
| 8 | Creating or maintaining documentation | 18.9 | 9.3% | 19.1% | 33.5% | Axis calculation on Stack Overflow 2025 |
| 9 | Testing code | 16.3 | 6.7% | 19.2% | 37.2% | Axis calculation on Stack Overflow 2025 |
| 10 | Committing and reviewing code | 11.8 | 3.8% | 15.9% | 49.7% | Axis calculation on Stack Overflow 2025 |
| 11 | Project planning | 10.1 | 4.1% | 12.1% | 59.0% | Axis calculation on Stack Overflow 2025 |
| 12 | Predictive analytics | 8.8 | 4.2% | 9.2% | 57.1% | Axis calculation on Stack Overflow 2025 |
| 13 | Deployment and monitoring | 6.0 | 2.3% | 7.4% | 64.8% | Axis calculation on Stack Overflow 2025 |
| Lifecycle SDAP (mean) | 19.6 | 38.3% | Axis Intelligence Research |
Inputs: Stack Overflow Developer Survey 2025, section 3.2, “AI in the development workflow,” retrieved September 15, 2026. Base sizes by response option: 11,202 / 20,991 / 22,518 / 12,790 / 25,349.
The delivery cliff
Group the thirteen stages into what engineers actually call them. Build-stage work (writing code, debugging, testing, documenting code) averages SDAP 22.2. Ship-and-govern work (committing and reviewing code, project planning, deployment and monitoring) averages SDAP 9.3.
According to Axis Intelligence Research, that ratio is 2.4 to 1. AI has entered the part of the pipeline where a mistake is caught by the next compile, and has been kept out of the part where a mistake reaches production.
The refusal numbers make the boundary explicit. 64.8% of developers say they do not plan to use AI for deployment and monitoring, 59.0% for project planning, 49.7% for committing and reviewing code. Compare that with 24.0% for writing code. Developers are not resisting AI. They are resisting AI at the merge button and the deploy button.
Sarah Mitchell: This is the finding that should reframe every enterprise AI rollout deck. The pitch is usually framed as end-to-end transformation of the SDLC. The data says developers have adopted AI at exactly the stages where the feedback loop is tight and reversible, and refused it where the blast radius is a production incident. That is not conservatism. That is correctly calibrated risk pricing by people who carry the pager.
Do Developers Trust AI-Generated Code?
Usage and trust are moving in opposite directions, and the divergence is now large enough to be structural rather than noise.
| Trust measure | Value | Population | Source |
|---|---|---|---|
| Highly trust AI output accuracy | 3.1% | All respondents | Stack Overflow 2025 |
| Somewhat trust | 29.6% | All respondents | Stack Overflow 2025 |
| Somewhat distrust | 26.1% | All respondents | Stack Overflow 2025 |
| Highly distrust | 19.6% | All respondents | Stack Overflow 2025 |
| Highly trust, 10+ years experience | 2.5% | Experienced developers | Stack Overflow 2025 |
| Highly trust, learning to code | 6.1% | Learners | Stack Overflow 2025 |
| Great deal or a lot of trust | 24% | Technology professionals | DORA 2025 |
| A little or no trust | 30% | Technology professionals | DORA 2025 |
Favorable sentiment toward AI tools fell to 59.7% in 2025 from above 70% in 2023 and 2024. Meanwhile 66% of developers report friction with AI output that is nearly correct but not quite, and 45.2% say debugging AI-generated code costs more time than writing it themselves. One in five, 20.0%, report reduced confidence in their own problem-solving.
Experience inverts trust cleanly. Learners are roughly 2.4 times more likely than ten-year veterans to report high trust in AI output. The people with production accountability trust the tools least, which maps precisely onto the delivery cliff in the SDAP table.
For tool-level trust and market-share detail, see our AI coding assistant statistics analysis and the GitHub Copilot statistics page.
How Many Developers Use AI Agents at Work?
Agents are the category everyone is forecasting and comparatively few are running.
30.9% of all Stack Overflow respondents use AI agents at work at any frequency: 14.1% daily, 9.0% weekly, 7.8% monthly or less. Against that, 37.9% say they do not plan to adopt agents, and a further 13.8% say they use AI exclusively in autocomplete or copilot mode.
Among developers who do run agents, the reported impact is individual rather than collective:
| Agent impact statement | Agree (strongly + somewhat) | Source |
|---|---|---|
| Reduced time on specific development tasks | 70.1% | Stack Overflow 2025 |
| Increased my productivity | 68.7% | Stack Overflow 2025 |
| Accelerated learning of new tech or codebases | 63.2% | Stack Overflow 2025 |
| Helped automate repetitive tasks | 64.2% | Stack Overflow 2025 |
| Solved complex problems more effectively | 49.0% | Stack Overflow 2025 |
| Improved the quality of my code | 37.5% | Stack Overflow 2025 |
| Improved collaboration within my team | 17.3% | Stack Overflow 2025 |
The spread between “increased my productivity” at 68.7% and “improved collaboration within my team” at 17.3% is 51.4 percentage points. According to Axis Intelligence Research, this is the clearest available evidence that agent gains are currently accruing to individuals and not to delivery organizations, which is the same conclusion DORA reaches from an entirely different dataset when it describes AI as an amplifier of whatever system it lands in.
Concerns scale with use. 86.9% of respondents agree they are concerned about agent output accuracy and 81.4% about data security and privacy. 28.2% report that internal IT or security policy blocks agent tools outright.
The agent tooling stack itself is being assembled from existing infrastructure rather than AI-native products. Among developers building with agents, Ollama leads orchestration at 51.1% and LangChain follows at 32.9%; for memory and data, Redis leads at 42.9% ahead of ChromaDB at 19.7% and pgvector at 17.9%; for observability, Grafana with Prometheus leads at 43.0% and Sentry at 31.8%. On the assistant side, ChatGPT reaches 81.7% and GitHub Copilot 67.9% among agent users.
Sarah Mitchell: Look at what developers picked for agent observability. Grafana, Prometheus, Sentry. Nothing on that list was designed for non-deterministic systems. Teams are monitoring agents with the same dashboards they use for stateless web services, which works right up until the failure mode is a plausible wrong answer rather than a crash. The observability gap is the next thing to break, and it will break quietly.
Which Programming Languages Are Winning in the AI Era?
Language share is now partly a function of how well a language constrains generated code. That is a genuinely new dynamic.
| Language | GitHub monthly contributors (Aug 2025) | YoY growth | New repos (Sep 2024 to Aug 2025) | Source |
|---|---|---|---|---|
| TypeScript | 2,636,006 | +66.63% | 5,394,256 | GitHub Octoverse 2025 |
| Python | ~2,600,000 | +48.78% | 9,261,587 | GitHub Octoverse 2025 |
| JavaScript | ~2,150,000 | +24.79% | 9,345,046 | GitHub Octoverse 2025 |
| Java | not published | +20.73% | 3,520,215 | GitHub Octoverse 2025 |
| C# | not published | +22.22% | 1,478,463 | GitHub Octoverse 2025 |
In August 2025 TypeScript overtook both Python and JavaScript on GitHub by distinct monthly contributors, ahead of Python by roughly 42,000 contributors. GitHub’s Octoverse 2025 attributes the shift to frameworks scaffolding in TypeScript by default and to type systems catching model-generated errors before they reach production.
Python holds the AI workload. It anchors 582,196 AI-tagged repositories, up 50.7% year over year, with Jupyter Notebook at 402,643 and TypeScript inside AI repositories at 85,746, up 77.9%. Shell scripting inside AI projects grew fastest at 324%, which is what eval harnesses and deployment glue look like in aggregate.
Self-reported primary languages tell a different story because the sampling frame differs. JetBrains finds Python at 35%, Java at 33%, JavaScript at 26%, TypeScript at 22% and HTML/CSS at 16% among 24,534 developers, with Go (11%), Rust (10%) and Python (7%) topping the list of languages developers want to adopt next.
Two measures, two answers. GitHub counts commit activity on a platform; JetBrains asks people what they mainly write. Axis Intelligence Research publishes both rather than averaging them, because a contributor count and a self-reported primary language are not the same quantity.
What Is Happening to Developer Jobs?
The federal projection moved, and the size of the move is the story.
| BLS release | Occupation scope | Period | Projected growth | Base employment | Source |
|---|---|---|---|---|---|
| Employment Projections 2023 to 2033 | Software developers | 2023 to 2033 | 17.9% | 1,692,100 | BLS |
| Occupational Outlook Handbook, Aug 2026 | Software developers | 2025 to 2035 | 10% | 1,717,800 | BLS |
| Occupational Outlook Handbook, Aug 2026 | QA analysts and testers | 2025 to 2035 | 6% | 187,600 | BLS |
| Occupational Outlook Handbook, Aug 2026 | All occupations | 2025 to 2035 | 3% | not applicable | BLS |
According to Axis Intelligence Research, the U.S. Bureau of Labor Statistics has revised its software developer growth projection down by 7.9 percentage points across two projection vintages, from 17.9% over 2023 to 2033 to 10% over 2025 to 2035. These are successive projection rounds with different base years and different decades, so the comparison is between vintages rather than a like-for-like restatement. It remains the largest downward revision to this occupation’s outlook in the AI period, and it happened while BLS still classified the growth rate as much faster than the average for all occupations.
The level numbers are less dramatic than the rate. Software developers held 1,717,800 jobs in 2025 with a projected 1,892,600 by 2035, and the combined developer, QA and testing group is projected to add 185,400 positions with about 106,100 annual openings once transfers and retirements are counted. Median annual wage for software developers was $135,980 in May 2025, against $50,980 for all occupations.
The entry-level squeeze shows up in survey data before it shows up in federal series. JetBrains found 61% of junior developers describe the job market as challenging versus 54% of senior developers, and 68% of all respondents expect employers to require AI proficiency.
Does AI Actually Make Developers Faster?
The measured evidence and the self-reported evidence still disagree, and the gap is narrowing rather than closing.
METR, a research nonprofit, ran the only randomized controlled trial on this question. In its early-2025 study, 16 experienced open-source developers working on their own repositories across 246 tasks took 19% longer with AI tools available, with a confidence interval of +2% to +39%. The same developers estimated afterwards that AI had made them 20% faster, a 39-percentage-point reversal in direction.
METR’s February 2026 follow-up matters more than the headline that made the rounds. In a second study of 57 developers across 143 repositories and more than 800 tasks, the subset of returning developers showed an estimated 18% speedup (confidence interval -38% to +9%), while newly recruited developers showed 4% (confidence interval -15% to +9%). METR published both numbers alongside a warning that it considers the experiment unreliable: 30% to 50% of developers said they withheld tasks they did not want to attempt without AI, and recruitment skewed against the most enthusiastic adopters. METR’s own reading is that the true effect is probably larger than measured and that the design needs replacing.
That is what honest measurement looks like, and it is why this article does not report a single productivity number.
Self-reports run far higher. DORA found more than 80% of respondents saying AI improved their productivity and 59% reporting a positive effect on code quality. METR’s own May 2026 survey of 349 technical workers found a self-reported median 1.4 to 2 times change in the value of work produced, published with explicit skepticism about the magnitude.
Meanwhile GitHub’s platform telemetry shows throughput rising: merged pull requests averaged 43.2 million monthly in 2025 against 35 million in 2024, code pushes 82.19 million against 65 million, and issues closed 4.25 million against 3.4 million. GitHub labels these observational signals rather than causal claims, and that framing is correct.
Sarah Mitchell: Three instruments, three answers, and the honest position is that the effect size is unresolved and probably task-dependent. What is not unresolved is the direction of the security signal. Broken access control overtook injection as the most common CodeQL alert on GitHub, flagged in more than 151,000 repositories, up 172% year over year, and GitHub links part of that to generated scaffolds that look correct and skip the auth check. Speed you can argue about. An endpoint with no authorization check is not an argument.
How Deep Is AI Integration Inside Teams?
Individual adoption and organizational integration are different measurements, and the gap between them is where AI budgets go to die.
- JetBrains: 85% of developers regularly use AI tools, while 62% rely on at least one AI coding assistant, agent or code editor. 15% have not adopted AI at all.
- Stack Overflow: 17.3% of agent users agree that agents improved team collaboration, the lowest-rated impact by a wide margin.
- DORA: 65% report heavy reliance on AI for software development (37% moderate, 20% a lot, 8% a great deal), and adoption is now linked to higher software delivery throughput, reversing the 2024 finding.
- DORA’s 2026 ROI follow-up describes returns arriving on a J-curve, with an initial negative phase before positive ROI, and states that positive ROI is not automatic.
DORA’s central conclusion from roughly 5,000 professionals is that AI functions as an amplifier: strong delivery systems get faster, weak ones get louder. That is consistent with what SDAP shows from a completely different survey. Teams that have not automated their review and deploy path cannot hand those stages to AI, so their penetration stays concentrated in the build stage, where individual speed gains accumulate and then queue behind an unchanged human review bottleneck.
For the broader market context around this spending, see our AI statistics report and the adoption trajectory in Cursor AI statistics.
Methodology
Collection. All figures were retrieved from primary publishers on September 15, 2026: Stack Overflow Developer Survey 2025 (sections on AI, technology and work), JetBrains State of Developer Ecosystem 2025, Google Cloud DORA State of AI-assisted Software Development 2025, GitHub Octoverse 2025, METR’s July 2025 randomized controlled trial and February 2026 experiment update, and two U.S. Bureau of Labor Statistics releases. No figure in this article is sourced from a secondary aggregator or from model memory. Every number appears as a row in the accompanying CSV with source organization, document, URL and retrieval date.
SDAP construction. Stack Overflow publishes its thirteen-task workflow matrix normalized within each of five response options rather than as all-respondent shares, with separate base sizes per option (11,202 / 20,991 / 22,518 / 12,790 / 25,349). Axis Intelligence Research reconstructed the all-respondent distribution as follows. For each task and each option, respondent count = published percentage × that option’s base size. For each task, all-respondent share = that task’s count for the option, divided by the sum of its counts across all five options.
Validation. Because each respondent assigns exactly one option per task, the reconstructed per-task totals should be approximately equal across all thirteen tasks. They are: reconstructed totals range from 29,098 to 30,485 with a mean of 29,961 and a maximum deviation of 2.9% from that mean. A reconstruction built on a mistaken reading of the normalization would not produce that convergence. The reconstructed totals also sit within the published response range for the question.
SDAP formula. SDAP = (share currently mostly AI × 1.0) + (share currently partially AI × 0.5), expressed 0 to 100. Weights are chosen so a stage fully handed to AI scores 100 and partial delegation counts half. Planned future use scores zero. Lifecycle SDAP is the unweighted mean of all thirteen stage scores, unweighted because no published data establishes the relative time share of each stage. Build-stage grouping is writing code, debugging or fixing code, testing code and documenting code. Ship-and-govern grouping is committing and reviewing code, project planning, and deployment and monitoring.
Scope of the SDAP reading. SDAP is computed on a single survey instrument with a self-selected international sample, so it measures the developer population that responds to Stack Overflow rather than all working developers. It is designed for annual recomputation against the same instrument so that the series stays internally comparable, and the reconstruction method is published here in full so any reader can reproduce the score from the public source.
Agent impact percentages are the sum of strongly agree and somewhat agree from Stack Overflow’s published Likert distributions, computed by Axis Intelligence Research and shown to one decimal place.
Source discrepancies published, not reconciled. Stack Overflow reports 84% adoption, JetBrains 85%, DORA 90%. JetBrains’ own blog reports 61% of junior versus 54% of senior developers describing the job market as challenging, while at least one trade report of the same study cites 34% for seniors; this article uses the publisher’s figure. GitHub contributor counts and JetBrains self-reported primary languages rank languages differently. None of these are averaged.
About This Dataset
Contents. 294 verified data rows covering developer AI adoption rates, daily usage frequency by experience level, thirteen-stage workflow penetration with the full SDAP calculation, trust distributions, agent adoption and impact, agent tooling stacks, programming language rankings, U.S. employment projections and wages, and controlled-trial productivity estimates. Temporal coverage: February 2025 through August 2026.
License. CC BY 4.0. Use, share and adapt with attribution.
Citation. Axis Intelligence Research, Developer AI Adoption Statistics 2026, 2026.
APA: Axis Intelligence Research, & Mitchell, S. (2026, September 15). Developer AI adoption statistics 2026: Workflow penetration, teams, languages and agents. Axis Intelligence. https://axis-intelligence.com/developer-ai-adoption-statistics/
MLA: Axis Intelligence Research and Sarah Mitchell. “Developer AI Adoption Statistics 2026: Workflow Penetration, Teams, Languages and Agents.” Axis Intelligence, 15 Sept. 2026, axis-intelligence.com/developer-ai-adoption-statistics/.
Chicago: Axis Intelligence Research and Sarah Mitchell. “Developer AI Adoption Statistics 2026: Workflow Penetration, Teams, Languages and Agents.” Axis Intelligence, September 15, 2026.
Frequently Asked Questions
What percentage of the software delivery lifecycle has AI actually taken over?
According to Axis Intelligence Research, the lifecycle-wide SDAP score is 19.6 out of 100 across thirteen stages. Penetration is highest for answer-seeking (39.3) and lowest for deployment and monitoring (6.0). The 84% headline adoption figure measures whether a developer touches AI at all, not how much of the pipeline runs on it.
Why do developers refuse to let AI near deployment and code review?
64.8% will not use AI for deployment and monitoring and 49.7% will not use it for committing and reviewing code, versus 24.0% for writing code. The pattern tracks reversibility: generated code that is wrong fails at compile or test, while a bad merge or deploy reaches users. GitHub’s own data shows broken access control alerts up 172% year over year across 151,000+ repositories, partly from generated scaffolds missing auth checks.
Do senior engineers trust AI-generated code less than juniors?
Yes, measurably. 2.5% of developers with ten or more years of experience report highly trusting AI output, against 6.1% of those learning to code. Experienced developers also post the highest rate of high distrust at 20.7%. Daily usage runs the same direction: 55.5% among early-career developers, 47.3% among experienced ones.
Is there any controlled evidence that AI coding tools make engineers faster?
The evidence is unsettled. METR’s randomized controlled trial of experienced open-source developers found tasks took 19% longer with AI in early 2025. Its February 2026 follow-up estimated an 18% speedup for returning participants and 4% for new ones, but METR published both with a warning that selection effects make the experiment unreliable and that the design is being replaced.
Why did TypeScript overtake Python and JavaScript on GitHub?
TypeScript reached 2,636,006 monthly contributors in August 2025, ahead of Python by about 42,000. GitHub attributes the shift to frameworks that scaffold TypeScript by default and to type systems catching model-generated errors before production, citing research that 94% of LLM-generated compilation errors were type-check failures.
How many developers run AI agents rather than autocomplete?
30.9% use agents at work at any frequency, with 14.1% daily. 13.8% explicitly say they use AI only in copilot or autocomplete mode, and 37.9% have no plans to adopt agents. Among agent users, 83.5% apply them to software engineering tasks.
Why do agent productivity gains not show up at team level?
68.7% of agent users agree agents increased their personal productivity, but only 17.3% agree agents improved team collaboration, a 51.4-point gap. DORA reaches a compatible conclusion from separate data: AI amplifies the delivery system it enters, so individual speed gains queue behind unchanged review and release processes rather than converting into throughput.
Has the U.S. government lowered its outlook for developer jobs because of AI?
BLS now projects 10% growth for software developers from 2025 to 2035, down from the 17.9% it projected for 2023 to 2033, a 7.9-percentage-point revision across two projection vintages. BLS still classifies the occupation as growing much faster than the 3% average, with about 106,100 annual openings and a May 2025 median wage of $135,980.
What share of developers still do not use AI tools at all?
16.2% of Stack Overflow respondents say they do not use AI and do not plan to, and JetBrains puts non-adopters at 15%. Among people learning to code the refusal rate is higher at 19.5%. Across the thirteen lifecycle stages, an average of 38.3% of developers say they do not plan to use AI for a given task.
How is SDAP calculated and can I reproduce it?
SDAP = (share currently mostly AI × 1.0) + (share currently partially AI × 0.5), scored 0 to 100 per lifecycle stage. Inputs come from Stack Overflow’s published workflow matrix, reconstructed to all-respondent shares using each response option’s base size. The full method, the validation check and every input value are in the Methodology section and in the downloadable CSV.