A $24 market, 14,495 postings, and three skills on the $30 line
Upwork's weekly job data puts the market at 14,495 postings for September 21 to 27, 2026, down 1% week over week with a $24 average advertised hour, and this issue picks the three skills whose advertised middle reaches the $30 line — API integrations, AI agent development, and Retool and internal tools — each with an honest 90-day learning path, three-tier pricing, client-screening red flags and adaptable gig copy.
The skill market this week
Upwork's job volume is now measured weekly, through the platform's own API. For the week of September 21 to 27, 2026, that count is 14,495 job posts, down 1% on the week before. The average advertised hourly rate was $24. Average fixed-price budgets rose 15.5%, to $604. 1
Twelve weeks of that series sit in a narrow band: 13,803 posts in the week of July 6, a high of 16,241 in mid-July, and nothing since outside 13,900 to 16,300. The average advertised hour moved once over those twelve weeks, from $23 to $24. 2
Where the money sits is narrower than the volume. Three of the twelve categories carry an average advertised hourly rate above $30: IT & Networking at $35, Data Science & Analytics at $31 and Legal at $57. Design & Creative is the largest category in the market, 4,414 posts and 31% of the week, and it averages $25 an hour. Customer Service averages $12 across 286 posts. Those are averages of the budgets clients stated; the median layer below comes from a second index. 1
Demand moved further than the headline. Comparing the week of August 24 to 30 with the week of September 21 to 27, Customer Service postings rose 15.8%, Accounting & Consulting 8.7% and IT & Networking 5.0%. Web, Mobile & Software Dev fell 11.1%, Data Science & Analytics 10.7% and Legal 8.9%. No category came near a 50% rise across those four weeks. 2
The rate layer works at skill level: a second index tracks 17 skill categories, on Upwork postings only, and reports medians. Over the 30 days ending September 28, 2026 it counted 7,578 Upwork posts that stated a budget. Of those, 4,301 were hourly, with a $23 median, a middle half of $12 to $35 and a top decile of $55. The other 3,277 were fixed-price, with a $200 median. Nine of the 17 categories post a median of $30 an hour or higher, and the highest is $35. 3
This channel screens for 50% demand growth inside 30 days plus a median above $30. The growth half has no skill-level public measurement; the finest published series is the weekly one above, and it covers 12 broad categories. Fiverr publishes no skill-level hourly median at all, because its catalogue quotes project packages: Pro sellers listed under API integration work start between $211 and $680 per project. 4 The rate column below is therefore Upwork-side, and the three picks land on or a dollar under the $30 line. Each pricing template is quoted against the quartiles those postings actually show, at the hours this work usually takes.
The three, side by side
Sorted by how many Upwork postings stated an hourly rate over the 30 days ending September 28, 2026, because that number is the sample standing behind every other figure in the row.
| Skill | 30-day demand | Median / top quartile | Best first buyer |
|---|---|---|---|
| API integrations | 452 hourly postings 5 | $30 / $43 | A small business whose two SaaS tools do not talk to each other |
| AI agent development | 186 hourly postings 6 | $29 / $45 | A team that wants one repetitive job handled end to end |
| Retool and internal tools | 27 hourly postings 7 | $30 / $48 | An operations team running its process on spreadsheets |
1. API integrations
Demand: 452 Upwork posts in this category stated an hourly rate over the 30 days ending September 28, 2026, the largest hourly sample in the index. 5 Across the five boards the same index tracks, API integration drew 1,631 postings in that window, the second-largest category on its board. 8 Clients describe a data-flow problem and seldom name a tool, which leaves the tool choice to you. 8
Rate reality: Median $30/hr, middle half $20 to $43, top quartile $43, top decile $70. Fixed-price projects are the larger half by count: 417 postings, a $250 median, a $1,000 top quartile and a $3,299 top decile. 5 These are client-stated budgets in USD over a rolling 30-day window, not money anyone collected.
Learning path: In three days, connect two services by hand: one authenticated request, the pagination, a rate limit, and the result written into a sheet. Read the endpoints in Postman before writing code, because a lot of client APIs arrive undocumented. 8 In four weeks, rebuild that connection as a scheduled job with retries, an alert when it fails and a handover note. Learn one platform properly: n8n when volume is high or the client will self-host, Make.com when they want to edit the flow themselves, Zapier when they already pay for it, and a small Node or Python service when the payloads need real validation. 8 In 12 weeks, carry one integration through a source system's change, whether that is a renamed field, a retired endpoint or a rotated key. That maintenance is what a retainer pays for.
Pricing template: Starter gig, $250 to $600 for one connection in one direction on a schedule, documented. Mid-tier package, $1,200 to $3,500 for several systems with error handling, retries and a handover file. Monthly retainer, $400 to $1,200 for monitoring, field changes and new connections.
Top 3 client-screening red flags: A post labelled "API integration" that describes one trigger and one action is a few-hundred-dollar Zapier job wearing an engineering title, and it builds no portfolio. A fixed-price post under $200 that describes OAuth flows, webhooks and retry logic sits below the floor its own scope requires; the index's own users filter at $500 fixed and $35 an hour. 9 A client who cannot supply API documentation or a test key before you quote is asking you to price reverse-engineering, and that reading time belongs in a separate discovery fee.
Sample Fiverr gig description:
I will connect the tools your business already pays for, so the data moves between them without anyone retyping it. You send the systems on each end, what should trigger the transfer, how often it should run, and which fields matter. I build the connection, handle the authentication, deal with pagination and rate limits, schedule the run, and set an alert so you hear about a failure before a report comes out wrong. The starter package covers one direction between two systems on a schedule. The standard package covers several systems with retries, error handling and a written handover, so someone else can maintain it. The retainer keeps it running when a source system renames a field or retires an endpoint. I work in n8n, Make.com, Zapier and Postman, and in Node.js or Python when payloads need real validation. Assume every connection needs maintenance; the retainer is what covers it, and I will tell you when a cheaper tool fits your volume better than the one you asked for. Best fit: small businesses and agencies running on two to five SaaS tools that do not talk to each other. Poor fit: a full application build, or a warehouse project inside the same fee. Send your tool list and the fields you need moved before ordering.
Adjacent specialization for the next $20/hr: Retrieval over the client's own documents, which the index files under RAG and knowledge bases. That category posts the index's highest median, $35/hr, with a top quartile of $50 across 48 hourly postings. 3 The client who needed two systems connected asks next for their contracts, tickets or manuals to be searchable, and the plumbing you built is most of the groundwork.
2. Retool and internal tools
Demand: 27 Upwork posts stated an hourly rate here over the 30 days ending September 28, 2026. That is the smallest sample of the three and it sits close to the index's publication floor of 25 postings, so treat it as evidence that the work exists. 7 Across the five boards, the category posted 92 jobs this month. 10
Rate reality: Median $30/hr, middle half $20 to $48, top quartile $48, top decile $75. Fixed-price postings are too few for the index to publish a median, so these hourly figures are the whole picture. 7
Learning path: In three days, build one internal screen against a sample database: a table, a search filter, a form that writes back, and one permission level. The SQL has to be real; the rest is a low-code builder. In four weeks, point it at a client's live system, add roles so a salesperson and a finance lead see different rows, and find out what breaks when two people edit the same record. In 12 weeks, retire one spreadsheet-and-email process at a small company and keep the replacement running for a month.
Pricing template: Starter gig, $400 to $900 for one screen over one data source. Mid-tier package, $1,500 to $4,000 for a multi-screen tool with roles, validation and alerts. Monthly retainer, $500 to $1,500 to extend and maintain it.
Top 3 client-screening red flags: A brief that requires the tool to live inside the client's own workspace, in their components and their naming, turns a build into repairing someone else's schema; price the reading time as its own line. A tool that will display customer records with no named owner for access control leaves a security decision the client has not made, and it lands on you. A fixed fee covering "a simple dashboard" will meet write-back, permissions and notifications in the second week.
Sample Fiverr gig description:
I will build the internal tool your team keeps rebuilding in a spreadsheet, so the work happens in one place with rules instead of a file everyone edits. You send the process as it runs today, the data behind it, who needs to see which rows, and what should happen when something is approved or rejected. I build the screens, connect your data, add the permissions, wire up the alerts, and hand over a short written guide so a new hire can use it. The starter package covers one screen over one data source. The standard package covers several screens with role-based access, validation rules and notifications. The retainer covers new screens, field changes and small fixes each month. I build in Retool and SQL, and I can work against Postgres, MySQL, Airtable or a REST API. Access control stays with you: I build the roles you specify, and I need a named person on your side who decides who gets them. Best fit: operations, finance and support teams running a recurring process through spreadsheets and email. Poor fit: a customer-facing product, or a rebuild of software you already pay for. Send your current process and your data sources before ordering.
Adjacent specialization for the next $20/hr: The data layer under the tool. Data pipelines and ETL posts a $33/hr median with a $45 top quartile across 55 hourly postings. 3 Once you maintain a tool across three systems, the client who needs those systems reconciled nightly is one conversation away.
3. AI agent development
Demand: 186 Upwork posts stated an hourly rate here over the 30 days ending September 28, 2026, the third-largest hourly sample in the index. 6 Across the five boards, the category posted 539 jobs this month. 10
Rate reality: Median $29/hr, a dollar under this channel's line, with a middle half of $20 to $45, a top quartile of $45 and a top decile of $60. Fixed-price work carries the ceiling: 108 postings, a $300 median, a $1,200 top quartile and a $7,000 top decile. 6 Upwork's own guide, which lists typical rates on freelancer profiles, puts AI developers at $30 to $50 an hour and AI engineers at $35 to $60, the band a practice reaches once it has work to show. 11
Learning path: In three days, build one agent with a bounded job: read a form, look up one record, write one reply, stop. Log every step. In four weeks, add the failure path, which is what the agent does when an API times out or the model returns something unusable, and connect one real system. In 12 weeks, run one agent in front of live traffic and own its evaluation: a fixed set of cases you re-run after every change, with the failures written down.
Pricing template: Starter gig, $300 to $700 for one agent on one task with logging. Mid-tier package, $1,200 to $3,500 for an agent with integrations, an escalation path and a documented set of test cases. Monthly retainer, $500 to $1,500 for monitoring, prompt changes and re-testing.
Top 3 client-screening red flags: A request to build the agent first and settle payment afterwards is an unpaid test with a longer deadline; the index puts the line at two hours and recommends a paid discovery session in place of it. 9 An offer of a share of what the agent saves or earns is commission work: you carry the delivery risk, the client keeps the asset, and the attribution is theirs to compute. 9 A brief whose measure of success is "handle our customer service", with no list of cases, no escalation rule and no volume figure, cannot be priced and will keep generating revisions.
Sample Fiverr gig description:
I will build one AI agent that takes over a single repetitive job inside your business and knows when to hand back to a person. You send the task as it is done today, the systems it touches, the answers it is allowed to give, the cases it must escalate, and how you want the work logged. I build the agent, connect it to your tools, write the escalation rules, test it against a set of real cases, and record a run so you can watch it work before it touches anything live. The starter package covers one agent on one task with logging. The standard package adds your integrations, an escalation path and a documented set of test cases. The retainer covers monitoring, prompt changes and re-testing each month. Agents still fail: unusual inputs, changed systems and ambiguous requests all produce bad output, which is why the escalation rules and the test cases are priced into the build. I run agents only on data your compliance owner has cleared for the purpose. Best fit: teams with one high-volume, rule-shaped task. Poor fit: an agent meant to replace a team whose process nobody has written down. Send the task and the cases before ordering.
Adjacent specialization for the next $20/hr: Running the agent over the client's own documents. The LangChain and LLM frameworks category carries a $53 top quartile across 51 hourly postings, and the framework layer is where retrieval, tool calls and evaluation sit. 3
How to choose
All three picks sit within a dollar of each other on the median and all three have a top quartile between $43 and $53, so rate makes a poor tiebreaker. Volume and specification separate them. API integrations has the most postings and the clearest brief, because a client describes a data flow and the tool choice is yours. AI agent development has the highest fixed-price ceiling and the hardest specification problem, because "an agent that handles this" has no scope yet. Retool and internal tools has the smallest sample, which is why its numbers carry a floor warning. Sell one contained package first, priced per deliverable, with the excluded work named in the gig text before the client orders.
Cover image: AI-generated illustration for this issue.
References
- 1Upwork Market Index: Weekly Freelance Report
snipework.com
- 2
- 3DevSnipe AI Automation Rate Index
devsnipe.com
- 4Hire the best API integration experts
fiverr.com
- 5API Integrations Freelance Rates
devsnipe.com
- 6AI Agent Development Freelance Rates
devsnipe.com
- 7Retool & Internal Tools Freelance Rates
devsnipe.com
- 8
- 9
- 10AI Automation Freelance Jobs by Category
devsnipe.com
- 11
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.