{"id":139,"date":"2026-07-24T07:11:17","date_gmt":"2026-07-24T07:11:17","guid":{"rendered":"https:\/\/mobilions.com\/blog\/?p=139"},"modified":"2026-07-24T07:12:45","modified_gmt":"2026-07-24T07:12:45","slug":"integrating-ai-into-human-workflows","status":"publish","type":"post","link":"https:\/\/mobilions.com\/blog\/integrating-ai-into-human-workflows\/","title":{"rendered":"Integrating AI Into Human Workflows: A Complete 2026 Guide"},"content":{"rendered":"\n

Here is the uncomfortable gap that defines integrating AI into human workflows<\/strong> in 2026. According to McKinsey\u2019s State of AI 2025<\/a>, 88 percent of organizations now use AI in at least one business function, up from 78 percent the year before, yet only 39 percent report any measurable impact on their bottom line, and most of those attribute less than 5 percent of profit to it. Nearly everyone has adopted AI. Almost no one is getting real value from it. The difference is not the model you pick. It is how you fit it into the way people actually work.<\/p>\n\n\n\n

I design and ship production AI systems, so let me give you the answer up front rather than burying it. The single strongest predictor of whether AI pays off, per that same McKinsey survey, is whether a company fundamentally redesigns the workflow around the AI instead of bolting the AI onto the old workflow. High performers are nearly three times as likely to have done exactly that. So this guide is not a list of AI tools. <\/p>\n\n\n\n

It is about the two decisions that actually determine success: where the AI goes in the workflow, and where the human stays. Get those right and the tool almost does not matter. Get them wrong and the best model in the world will sit unused.<\/p>\n\n\n\n

Key takeaways<\/a><\/h2>\n\n\n\n

If you only have a minute, these are the points that matter most about integrating AI into human workflows.<\/p>\n\n\n\n

Adoption is not the problem, impact is. 88 percent of organizations use AI, but only 39 percent see bottom-line impact, and the gap is almost entirely about workflow design, not model choice.<\/p>\n\n\n\n

Redesign the workflow, do not bolt AI on. McKinsey found that fundamentally redesigning the workflow is the strongest single predictor of AI impact, and high performers are about three times as likely to do it.<\/p>\n\n\n\n

Decide augment versus automate for each task, not for the whole job. Most real gains come from AI augmenting a person on the parts it is good at, while the person keeps the judgment, not from replacing the person outright.<\/p>\n\n\n\n

Keep a human in the loop where it counts. Put people at the decision points that carry risk, nuance, or accountability, and let the AI run the rest. Gartner projects that over 40 percent of agentic AI projects will be scrapped by 2027, largely from missing exactly this.<\/p>\n\n\n\n

Why do most AI-in-workflow efforts stall?<\/a><\/h2>\n\n\n\n

The most common failure is not technical. A team buys an AI tool or wires up a model, drops it next to an existing process, and expects the process to get faster on its own. It rarely does, because the old process was designed around human constraints that no longer apply, and it still contains all the handoffs, approvals, and manual steps that made sense before AI existed. Adding AI to a workflow built for humans just gives you a human workflow with an AI bolted to the side.<\/p>\n\n\n\n

McKinsey\u2019s data makes this concrete. Roughly two-thirds of organizations have not yet scaled AI beyond experiments, and only 39 percent see any profit impact. Meanwhile the small group of high performers, about 6 percent of respondents, do a specific thing differently: they are nearly three times as likely to have fundamentally redesigned their workflows, three times as likely to have senior leaders actually own the AI effort, and three times as likely to pursue transformative change rather than small efficiency wins. The pattern is clear. Value comes from rethinking the work, not from sprinkling AI on top of it.<\/p>\n\n\n\n

This is why integrating AI into human workflows is a design problem before it is an engineering problem. Before anyone writes a prompt or picks a model, someone has to look at the actual work, decide which parts AI should do, which parts a person must keep, and how the two hand off to each other. That redesign is the project. The model is just a component.<\/p>\n\n\n\n

Should you augment or automate? The first real decision<\/a><\/h2>\n\n\n\n

The biggest conceptual mistake is treating AI integration as an all-or-nothing automation question. In practice, the useful unit is the task, not the job. A single person\u2019s role is made of dozens of tasks, and AI is excellent at some of them, mediocre at others, and dangerous at a few. The job is to sort them.<\/p>\n\n\n\n

\"Automate<\/figure>\n\n\n\n

Automate<\/strong> the tasks that are repetitive, high-volume, rule-based, and low-risk when they occasionally go wrong: sorting tickets, extracting data from documents, drafting first-pass summaries, categorizing inbound requests. Augment<\/strong> the tasks where a human brings judgment, context, or accountability but AI can do the heavy lifting underneath: a support agent who lets AI draft the reply but edits and sends it, an analyst who lets AI pull and structure the data but decides what it means, a lawyer who lets AI find the relevant clauses but makes the call. Keep fully human<\/strong> the tasks that carry real consequence, need empathy, or require someone accountable to stand behind them.<\/p>\n\n\n\n

Approach<\/td>Best for<\/td>Example<\/td>Who is in charge<\/td><\/tr><\/thead>
Automate<\/td>Repetitive, high-volume, rule-based, low-risk tasks<\/td>Categorizing tickets, extracting data from documents<\/td>The AI, running unattended<\/td><\/tr>
Augment<\/td>Judgment tasks where AI can do the heavy lifting underneath<\/td>Drafting a reply a person edits and sends<\/td>The human, with AI assisting<\/td><\/tr>
Keep human<\/td>Tasks with real consequence, empathy, or accountability<\/td>Handling an angry enterprise customer<\/td>The human, fully<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

The reason this matters is that augmentation, not automation, is where most of the near-term value actually lives. It keeps the human judgment that AI still lacks while removing the drudgery that wastes that judgment. When we scope AI work with clients through our AI development<\/a> practice, this task-by-task sort is almost always the first exercise, because it decides the entire shape of what gets built.<\/p>\n\n\n\n

Where does the human go? Human-in-the-loop patterns<\/a><\/h2>\n\n\n\n
\"Automate<\/figure>\n\n\n\n

Once you know which tasks are augmented rather than fully automated, the next question is exactly where the human sits in the flow. This is what people mean by human-in-the-loop, and it is not one thing. There are a handful of proven patterns, well summarized in Zapier\u2019s breakdown of human-in-the-loop<\/a>, and picking the right one per step is most of the design work.<\/p>\n\n\n\n

An approval flow<\/strong> pauses the workflow at a checkpoint so a person can approve, reject, or edit the AI\u2019s output before it proceeds, which is the right pattern when the action is visible to a customer or hard to undo. Confidence-based routing<\/strong> lets the AI act on its own when it is sure and escalate to a human only when its confidence drops below a threshold, which concentrates human attention exactly where the AI is shaky.<\/p>\n\n\n\n

Escalation paths<\/strong> send anything outside the AI\u2019s scope, such as a refund above a set value, to the right person instead of forcing the automation to guess. Feedback loops<\/strong> let humans correct AI outputs in a way that becomes training data, so the system improves over time. And audit logging<\/strong> records every automated action for later review without slowing anything down, which gives you traceability even on the steps that run unattended.<\/p>\n\n\n\n

Pattern<\/td>What it does<\/td>Use it when<\/td><\/tr><\/thead>
Approval flow<\/td>Pauses for a person to approve, reject, or edit before proceeding<\/td>The action is customer-visible or hard to undo<\/td><\/tr>
Confidence-based routing<\/td>AI acts when sure, escalates to a human when uncertain<\/td>You want human attention only where the AI is shaky<\/td><\/tr>
Escalation path<\/td>Routes out-of-scope cases to the right person<\/td>A request crosses a threshold, such as refund value<\/td><\/tr>
Feedback loop<\/td>Turns human corrections into training data<\/td>You want the system to improve over time<\/td><\/tr>
Audit logging<\/td>Records every automated action for later review<\/td>You need traceability on unattended steps<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

The skill is not using all of these everywhere. It is putting a human in the loop where decisions carry risk, nuance, or accountability, and letting the AI run unattended everywhere else. Put a person on every step and you have not saved anyone any time. Put a person on no steps and you get the failure mode the next section is about.<\/p>\n\n\n\n

How do you actually redesign a workflow around AI?<\/a><\/h2>\n\n\n\n

Redesign sounds abstract, so here is the concrete version I use. Start by mapping the current workflow as it really runs, every step, handoff, and decision, not the idealized version in a process doc. Then, for each step, sort it into automate, augment, or keep-human using the test above.<\/p>\n\n\n\n

Now comes the part teams skip: redraw the workflow assuming the automated and augmented steps are nearly instant and nearly free. Handoffs that existed only because a human step was slow can often disappear. Approvals that existed only to catch human error may move to a confidence threshold. The shape of the new workflow is usually different from the old one, and that difference is where the McKinsey impact comes from.<\/p>\n\n\n\n

Then design the human-in-the-loop points deliberately using the patterns above, instrument everything so you can measure it, and roll it out to a small slice of real work before scaling. The order matters. Most failed integrations map the workflow, add AI to each step in place, and stop, which is the bolting-on trap. The redesign step, redrawing the flow around what AI makes cheap, is the one that actually moves the numbers, and it is usually where a custom software<\/a> build is required, because off-the-shelf tools assume the old shape of the work.<\/p>\n\n\n\n

What about agentic AI, where the AI runs multiple steps itself?<\/a><\/h2>\n\n\n\n

The frontier of integrating AI into human workflows in 2026 is agentic AI, where instead of assisting one step, an AI agent plans and executes a sequence of steps on its own. McKinsey found 62 percent of organizations are at least experimenting with agents and 23 percent are scaling them somewhere, so this is real and moving fast. It is also where the human-in-the-loop question gets sharpest, because an agent taking ten actions unattended can go ten steps wrong before anyone notices.<\/p>\n\n\n\n

The honest data is sobering. Gartner projects that over 40 percent of agentic AI projects will be scrapped by the end of 2027<\/a>, citing escalating costs, unclear value, and inadequate risk controls, and independent coverage keeps arriving at the same conclusion: AI agents fail without human oversight<\/a>.<\/p>\n\n\n\n

This is not an argument against agents. It is an argument for designing them the same way as any other AI integration: give the agent the steps it can run unattended, put approval and confidence checkpoints at the consequential moments, log everything, and keep a person accountable for the outcome. The teams that treat agents as fully autonomous employees are the ones filling out that 40 percent. The teams that treat them as fast, tireless workers who still report to a human are the ones getting value.<\/p>\n\n\n\n

How do you know if it is working?<\/a><\/h2>\n\n\n\n

You measure it, and you measure the right thing. The trap is measuring adoption, how many people use the tool, when what matters is impact, whether the work is actually better, faster, or cheaper with quality holding. Pick a baseline before you start: how long the task takes, the error rate, the cost per unit, the throughput. <\/p>\n\n\n\n

Then compare honestly after, and watch for the quiet failure where AI makes a step faster but pushes errors downstream so the total workflow is no better. McKinsey\u2019s whole adoption-to-impact gap is really a measurement story: plenty of usage, little proven value, because few teams instrumented the workflow well enough to know. If you cannot state the before-and-after number for the workflow you changed, you have adopted AI but you have not yet integrated it.<\/p>\n\n\n\n

How do you get people to actually adopt it?<\/a><\/h2>\n\n\n\n

Here is the part that is easy to underrate: the hardest problem in integrating AI into human workflows is usually not the AI, it is the humans. A redesigned workflow only delivers value if the people in it trust it and use it, and trust is not automatic. People who feel the AI was dropped on them to replace them will quietly route around it, and a workflow everyone works around is worse than the one you had. The teams that succeed treat adoption as part of the design, not an afterthought.<\/p>\n\n\n\n

In practice that means a few things. Involve the people who do the work in the redesign, because they know where the real friction is and they adopt what they helped build. Be explicit that augmentation is removing their drudgery, not their job, and then make sure that is actually true. Start where the pain is obvious so the first win is felt, not argued. <\/p>\n\n\n\n

And give people an easy way to correct the AI and see their corrections matter, which is exactly what the feedback-loop pattern is for. This is the same reason we lean on genuinely useful, well-scoped tools rather than flashy ones, the way we approach the real-time and applied AI systems<\/a> we build. The best-designed workflow on paper still fails if the people in it do not believe in it.<\/p>\n\n\n\n

A real-world scenario<\/a><\/h2>\n\n\n\n

To make this concrete, picture a mid-size company\u2019s customer support team drowning in inbound tickets. The tempting move is to buy an AI chatbot and point it at the queue. The redesign move is different.<\/p>\n\n\n\n

A systematic reading of this guide sorts it quickly. First, map the real workflow: tickets arrive, get categorized, get researched, get answered, and some get escalated. Then sort each step. Categorizing tickets is repetitive and low-risk, so automate it. Drafting the answer is where AI does the heavy lifting but a human should still approve customer-facing replies, so augment it with an approval flow. Judging an angry enterprise customer who is threatening to churn needs empathy and accountability, so keep it human, routed by an escalation path. <\/p>\n\n\n\n

Add confidence-based routing so the AI answers the easy, high-confidence tickets end to end and sends the ambiguous ones to a person. Log everything for later review. The result is not a chatbot bolted onto the old queue. It is a redesigned workflow where AI handles volume, humans handle judgment, and the handoffs are deliberate. That is the version that actually cuts response time without wrecking customer trust.<\/p>\n\n\n\n

Myths and common mistakes<\/a><\/h2>\n\n\n\n

A few misconceptions cause most of the wasted effort.<\/p>\n\n\n\n

The first myth is that integrating AI means automating jobs. It almost never does at first. It means automating and augmenting tasks, and the biggest early wins are augmentation, where a person stays in charge.<\/p>\n\n\n\n

The second mistake is bolting AI onto the existing process. If you do not redesign the workflow, you keep all the handoffs and approvals built for a slower, human-only world, and you cap your upside at a small efficiency gain. This is the single most common reason AI projects underdeliver.<\/p>\n\n\n\n

The third mistake is going fully autonomous too early, especially with agents. Removing the human from consequential decisions is how you end up in Gartner\u2019s 40 percent that get scrapped. Autonomy is earned step by step as the system proves itself, not granted on day one.<\/p>\n\n\n\n

The fourth mistake is measuring adoption instead of impact. Lots of logins is not value. If you cannot show the workflow got measurably better, the integration is not done.<\/p>\n\n\n\n

The honest caveat worth stating plainly: this is genuinely hard, and it is more organizational than technical. The models are capable enough today. The bottleneck is redesigning how people work and getting them to trust and adopt the new flow, which is change management as much as engineering. Any guide that makes it sound like a plug-in is selling you the easy 20 percent and skipping the 80 that decides the outcome.<\/p>\n\n\n\n

Why Mobilions<\/a><\/h2>\n\n\n\n

Mobilions has been building custom software, mobile apps, and AI solutions since 2016. We have delivered more than 250 projects for over 100 clients across 20-plus countries, which means we have integrated AI into real human workflows, not just demoed models. When we take on this work, we start with the task-by-task sort and the workflow redesign rather than the model, we design the human-in-the-loop points deliberately, and we instrument the workflow so you can actually prove the impact.<\/p>\n\n\n\n

If you are planning to integrate AI into how your team works and want to get the design right before writing code, that is the conversation our AI development<\/a> team has with leaders every week, and where it helps we pair it with the engineers who have shipped<\/a> these systems before.<\/p>\n\n\n\n

Summary<\/a><\/h2>\n\n\n\n

Integrating AI into human workflows in 2026 is not a tooling problem, it is a design problem. Adoption is nearly universal at 88 percent of organizations, but only 39 percent see real impact, and the difference is workflow redesign, the strongest predictor McKinsey found.<\/p>\n\n\n\n

Sort the work task by task into automate, augment, and keep-human. Put humans in the loop at the points that carry risk, nuance, or accountability using proven patterns like approval flows and confidence-based routing, and let AI run the rest. Redesign the flow around what AI makes cheap rather than bolting AI onto the old process. Be especially careful with agents, since over 40 percent of agentic projects are projected to fail, almost always from removing human oversight too soon. Measure impact, not adoption. Get the design right and the model is the easy part.<\/p>\n\n\n\n

Frequently asked questions<\/a><\/h2>\n\n\n
\n
\n
\n

What does integrating AI into human workflows actually mean?<\/strong><\/h3>\n
\n\n

It means redesigning how work gets done so AI and people each handle the parts they are best at, with deliberate handoffs between them. It is not just adding an AI tool to an existing process. The work is deciding which tasks AI should automate, which it should augment with a human in charge, and which stay fully human.<\/p>\n\n<\/div>\n<\/div>\n

\n

Why do so many AI workflow projects fail to deliver value?<\/strong> <\/h3>\n
\n\n

Because most teams bolt AI onto their existing process instead of redesigning it. McKinsey found that 88 percent of organizations use AI but only 39 percent see bottom-line impact, and the strongest predictor of impact is fundamentally redesigning the workflow, which most teams skip.<\/p>\n\n<\/div>\n<\/div>\n

\n

What is the difference between augmenting and automating with AI?<\/strong> <\/h3>\n
\n\n

Automating means AI does a task end to end without a person, which suits repetitive, low-risk, rule-based work. Augmenting means AI does the heavy lifting while a human keeps judgment and accountability, such as drafting a reply the person edits and sends. Most early value comes from augmentation, not full automation.<\/p>\n\n<\/div>\n<\/div>\n

\n

What is human-in-the-loop and when should you use it?<\/strong><\/h3>\n
\n\n

Human-in-the-loop means placing people at specific decision points in an otherwise automated workflow. Use it where decisions carry risk, nuance, compliance implications, or need accountability. Common patterns include approval flows, confidence-based routing that escalates only uncertain cases, escalation paths, feedback loops, and audit logging.<\/p>\n\n<\/div>\n<\/div>\n

\n

How do you redesign a workflow around AI?<\/strong><\/h3>\n
\n\n

Map the current workflow step by step, sort each step into automate, augment, or keep-human, then redraw the flow assuming the automated and augmented steps are nearly instant, which often removes handoffs and approvals that only existed because human steps were slow. Then design the human-in-the-loop points, instrument everything, and roll out to a small slice before scaling.<\/p>\n\n<\/div>\n<\/div>\n

\n

Is agentic AI safe to put in production workflows<\/strong><\/h3>\n
\n\n

?<\/strong> It can be, but only with human oversight designed in. Gartner projects over 40 percent of agentic AI projects will be scrapped by 2027, largely from inadequate controls. The safe pattern is to let an agent run the steps it can handle unattended while keeping approval and confidence checkpoints at consequential moments and a person accountable for the outcome.<\/p>\n\n<\/div>\n<\/div>\n

\n

How do you measure whether an AI workflow integration is working?<\/strong><\/h3>\n
\n\n

Measure impact, not adoption. Set a baseline before you start, such as task time, error rate, cost per unit, and throughput, then compare honestly after, watching for cases where a step gets faster but pushes errors downstream. If you cannot state the before-and-after number for the workflow, the integration is not finished.<\/p>\n\n<\/div>\n<\/div>\n

\n

Will integrating AI replace my employees?<\/strong><\/h3>\n
\n\n

Usually not, at least not first. The useful unit is the task, not the job, and most roles are a mix of tasks where AI augments the person rather than replacing them. The near-term pattern is people doing more valuable work because AI removed the drudgery, not people being removed.<\/p>\n\n<\/div>\n<\/div>\n

\n

Where should a company start with integrating AI into workflows?<\/strong> <\/h3>\n
\n\n

Start with one real workflow that has clear, measurable pain, map it honestly, sort its tasks into automate, augment, and keep-human, redesign the flow, add deliberate human-in-the-loop checkpoints, and measure the before and after. A focused, measured pilot beats a broad rollout of AI tools that never gets redesigned into the work.<\/p>\n\n<\/div>\n<\/div>\n

\n

Which tasks should you automate with AI first?<\/strong><\/h3>\n
\n\n

Start with tasks that are repetitive, high-volume, rule-based, and low-risk when they occasionally go wrong, like categorizing tickets, extracting data from documents, or drafting first-pass summaries. Leave judgment, empathy, and accountability tasks to people. The useful unit is the task, not the whole job, so sort each one before automating anything.<\/p>\n\n<\/div>\n<\/div>\n

\n

What are the biggest challenges of integrating AI into workflows?<\/strong><\/h3>\n
\n\n

The hardest parts are organizational, not technical. Teams bolt AI onto an old process instead of redesigning it, remove human oversight too early, measure adoption instead of impact, and underestimate change management. The models are usually capable enough already. The real bottleneck is redesigning how people work and getting them to trust the new flow.<\/p>\n\n<\/div>\n<\/div>\n

\n

How do you get your team to adopt AI tools?<\/strong><\/h3>\n
\n\n

Treat adoption as part of the design, not an afterthought. Involve the people who do the work in the redesign, be explicit that augmentation removes drudgery rather than jobs, start where the pain is obvious so the first win is felt, and give people an easy way to correct the AI and see their corrections actually matter.<\/p>\n\n<\/div>\n<\/div>\n

\n

How much training do employees need to use AI tools?<\/strong><\/h3>\n
\n\n

Less than teams expect when the AI is designed into the workflow well, and more when it is bolted on awkwardly. The goal is tools that fit how people already work, so training focuses on the new handoffs and when to trust or override the AI, rather than on operating a complex separate system.<\/p>\n\n<\/div>\n<\/div>\n

\n

How do you choose the right AI tool for a workflow?<\/strong><\/h3>\n
\n\n

Decide the workflow redesign first, then pick a tool that fits it, not the other way around. Ask how it handles human-in-the-loop checkpoints, how it logs and audits actions, how it improves from corrections, and how it fits your existing systems. A tool that forces you back into the old process shape is the wrong tool.<\/p>\n\n<\/div>\n<\/div>\n

\n

How much does it cost to integrate AI into a workflow?<\/strong><\/h3>\n
\n\n

It varies with scope, but the largest cost is usually the redesign and integration work, not the model or tool license. A focused pilot on one workflow is comparatively cheap and is the right way to prove impact before spending more. Watch for hidden costs in agentic projects, which Gartner links to many being scrapped.<\/p>\n\n<\/div>\n<\/div>\n

\n

Why do companies abandon their AI tools?<\/strong> <\/h3>\n
\n\n

Usually because the tool was dropped onto an unchanged process, delivered no measurable impact, and lost the trust of the people meant to use it. Gartner projects over 40 percent of agentic AI projects will be scrapped by 2027, citing rising cost, unclear value, and weak controls. Abandonment is a design and adoption failure, not a model failure.<\/p>\n\n<\/div>\n<\/div>\n<\/div>\n<\/div>","protected":false},"excerpt":{"rendered":"

Here is the uncomfortable gap that defines integrating AI into human workflows in 2026. According to McKinsey\u2019s State of AI 2025, 88 percent of organizations now use AI in at least one business function, up from 78 percent the year before, yet only 39 percent report any measurable impact on their bottom line, and most […]<\/p>\n","protected":false},"author":6,"featured_media":144,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[11],"tags":[60,62,57,59,58,61],"class_list":["post-139","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai","tag-agentic-ai","tag-ai-adoption","tag-ai-integration","tag-ai-workflows","tag-human-in-the-loop","tag-workflow-automation"],"_links":{"self":[{"href":"https:\/\/mobilions.com\/blog\/wp-json\/wp\/v2\/posts\/139","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/mobilions.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/mobilions.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/mobilions.com\/blog\/wp-json\/wp\/v2\/users\/6"}],"replies":[{"embeddable":true,"href":"https:\/\/mobilions.com\/blog\/wp-json\/wp\/v2\/comments?post=139"}],"version-history":[{"count":2,"href":"https:\/\/mobilions.com\/blog\/wp-json\/wp\/v2\/posts\/139\/revisions"}],"predecessor-version":[{"id":145,"href":"https:\/\/mobilions.com\/blog\/wp-json\/wp\/v2\/posts\/139\/revisions\/145"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/mobilions.com\/blog\/wp-json\/wp\/v2\/media\/144"}],"wp:attachment":[{"href":"https:\/\/mobilions.com\/blog\/wp-json\/wp\/v2\/media?parent=139"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/mobilions.com\/blog\/wp-json\/wp\/v2\/categories?post=139"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/mobilions.com\/blog\/wp-json\/wp\/v2\/tags?post=139"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}